Loading [MathJax]/jax/element/mml/optable/BasicLatin.js
Research article Special Issues

FOA-BDNet: A behavior detection algorithm for elevator maintenance personnel based on first-order deep network architecture

  • Received: 24 July 2024 Revised: 28 September 2024 Accepted: 24 October 2024 Published: 04 November 2024
  • MSC : 68T07, 68T45, 74A08

  • The operation space of the vertical lift shaft is small, the components are complex, the occluding and different behavior space characteristics are similar, and the unsafe behavior is not easy to detect, which makes the operation safety of maintenance personnel in the elevator greatly threatened. This paper proposes an elevator maintenance personnel behavior detection algorithm based on the first-order deep network architecture (FOA-BDNet). First, a lightweight backbone feature extraction network is designed to meet the online real-time requirements of elevator maintenance environment monitoring video stream detection. Then, the feature fusion network structure of "far intersection and close connection" is proposed to fuse the fine-grained information with the coarse-grained information and to enhance the expression ability of deep semantic features. Finally, a first-order deep target detection algorithm adapted to the elevator scene is designed to identify and locate the behavior of maintenance personnel and to correctly detect unsafe behaviors. Experiments show that the detection accuracy rate on the self-built data set in this paper is 98.68%, which is 4.41% higher than that of the latest target detection model YOLOv8-s, and the reasoning speed reaches 69.51fps/s, which can be easily deployed in common edge devices and meet the real-time detection requirements for the unsafe behaviors of elevator scene maintenance personnel.

    Citation: Zengming Feng, Tingwen Cao. FOA-BDNet: A behavior detection algorithm for elevator maintenance personnel based on first-order deep network architecture[J]. AIMS Mathematics, 2024, 9(11): 31295-31316. doi: 10.3934/math.20241509

    Related Papers:

    [1] Mahmoud A. E. Abdelrahman, Wael W. Mohammed, Meshari Alesemi, Sahar Albosaily . The effect of multiplicative noise on the exact solutions of nonlinear Schrödinger equation. AIMS Mathematics, 2021, 6(3): 2970-2980. doi: 10.3934/math.2021180
    [2] Jalil Manafian, Onur Alp Ilhan, Sizar Abid Mohammed . Forming localized waves of the nonlinearity of the DNA dynamics arising in oscillator-chain of Peyrard-Bishop model. AIMS Mathematics, 2020, 5(3): 2461-2483. doi: 10.3934/math.2020163
    [3] Yongyi Gu, Najva Aminakbari . Two different systematic methods for constructing meromorphic exact solutions to the KdV-Sawada-Kotera equation. AIMS Mathematics, 2020, 5(4): 3990-4010. doi: 10.3934/math.2020257
    [4] P. Veeresha, D. G. Prakasha, Jagdev Singh . Solution for fractional forced KdV equation using fractional natural decomposition method. AIMS Mathematics, 2020, 5(2): 798-810. doi: 10.3934/math.2020054
    [5] M. Hafiz Uddin, M. Ali Akbar, Md. Ashrafuzzaman Khan, Md. Abdul Haque . New exact solitary wave solutions to the space-time fractional differential equations with conformable derivative. AIMS Mathematics, 2019, 4(2): 199-214. doi: 10.3934/math.2019.2.199
    [6] Yunmei Zhao, Yinghui He, Huizhang Yang . The two variable (φ/φ, 1/φ)-expansion method for solving the time-fractional partial differential equations. AIMS Mathematics, 2020, 5(5): 4121-4135. doi: 10.3934/math.2020264
    [7] Yousef Jawarneh, Humaira Yasmin, Ali M. Mahnashi . A new solitary wave solution of the fractional phenomena Bogoyavlenskii equation via Bäcklund transformation. AIMS Mathematics, 2024, 9(12): 35308-35325. doi: 10.3934/math.20241678
    [8] Weiping Gao, Yanxia Hu . The exact traveling wave solutions of a class of generalized Black-Scholes equation. AIMS Mathematics, 2017, 2(3): 385-399. doi: 10.3934/Math.2017.3.385
    [9] S. Owyed, M. A. Abdou, A. Abdel-Aty, H. Dutta . Optical solitons solutions for perturbed time fractional nonlinear Schrodinger equation via two strategic algorithms. AIMS Mathematics, 2020, 5(3): 2057-2070. doi: 10.3934/math.2020136
    [10] Safoura Rezaei Aderyani, Reza Saadati, Javad Vahidi, Nabil Mlaiki, Thabet Abdeljawad . The exact solutions of conformable time-fractional modified nonlinear Schrödinger equation by Direct algebraic method and Sine-Gordon expansion method. AIMS Mathematics, 2022, 7(6): 10807-10827. doi: 10.3934/math.2022604
  • The operation space of the vertical lift shaft is small, the components are complex, the occluding and different behavior space characteristics are similar, and the unsafe behavior is not easy to detect, which makes the operation safety of maintenance personnel in the elevator greatly threatened. This paper proposes an elevator maintenance personnel behavior detection algorithm based on the first-order deep network architecture (FOA-BDNet). First, a lightweight backbone feature extraction network is designed to meet the online real-time requirements of elevator maintenance environment monitoring video stream detection. Then, the feature fusion network structure of "far intersection and close connection" is proposed to fuse the fine-grained information with the coarse-grained information and to enhance the expression ability of deep semantic features. Finally, a first-order deep target detection algorithm adapted to the elevator scene is designed to identify and locate the behavior of maintenance personnel and to correctly detect unsafe behaviors. Experiments show that the detection accuracy rate on the self-built data set in this paper is 98.68%, which is 4.41% higher than that of the latest target detection model YOLOv8-s, and the reasoning speed reaches 69.51fps/s, which can be easily deployed in common edge devices and meet the real-time detection requirements for the unsafe behaviors of elevator scene maintenance personnel.



    This work deals with the numerical and analytical treatment of a second-kind delay Volterra integral equation (DVIE)

    x(T)=g(T)+T0B1(T,τ)x(τ)dτ+qT0B2(T,τ)x(τ)dτ,TJ:=[0,T], (1.1)

    where g(T), B1(T,τ), and B2(T,τ) are smooth enough on their respective domains in terms of J, D1:={(T,τ):0τTT}, and D2:={(T,τ):0τqTT}, so that based on [2], Eq (1.1) can be solved for a unique solution x(T), with respect to the variable vanishing delay of qT with 0<q<1. The reason for calling delay function qT with vanishing delay term more clearly turns back to that, it vanishes at the beginning point of the interval J. Moreover, X:=CM+R(J:=[0,T]) is a Banach space, where x(T), DVIE (1.1) solution lies as long as for the given data function g(T), we have gCM+R(J:=[0,T]), too. In more details, M stands for number of used collocation parameters, and R illustrates the first R start solutions obtained via the classical collocation scheme in the multistep collocation strategy as well.

    By virtue of Volterra integral equations and also their functional counterparts along with delay terms, whether vanishing or non-vanishing, a wide spectrum of science subjects, namely biology, ecology, physics and chemistry, have been mathematically well-formulated in order to analyse and study underlying phenomena. Particularly, these classes of mathematical modellings can be found in fluid dynamics, viscoelasticity of materials, population growth dynamics, heat conduction, epidemiology, controlled liquidation in obsolete production units, and renovation in economic systems; see [3,4,5,6,7] and references therein.

    Taking a look at the works being furnished to solve (1.1) results in constructing fast rate convergent multistep collocation-based strategies. This issue emerges because of the fact that, for example, [8] illustrates a global convergence order of M+1 for under study equation (1.1) on uniform meshes, while when it comes to superconvergence orders, it is notable that the order is at most 2M, most essentially with q=1/2, even M, as well as the M Gauss-Legendre collocation points. It is worth remarking that Brunner et al. [9] explored geometric meshes for solving (1.1), which yields to reach again a global convergence order of M+1 for iterated collocation methods with the superconvergence order 2MϵN, by regarding hypotheses of ϵN0 with N.

    Seemingly, the concept of exploiting non-uniform meshes makes sense to obtain higher superconvergence orders, most specifically for vanishing delay VIEs of the second type like (1.1). Pursuing this goal, Ming et al. [1] utilized quasi-geometric meshes in order to gain 2M superconvergence rates for the under consideration second-kind VIEs of pantograph type. As shown, there, quasi-geometric collocation schemes come into M+1 globally convergent numerical approximations, while in this study, we have reached the same or even higher superconvergence orders with a great deal of convenience and pace.

    For these reasons, and by following our previous works [10,11,12], through this paper, we aim to develop multistep collocation techniques for second-type VIEs with vanishing delays (1.1) on uniform meshes, in order to improve global convergence rate up to M+R with R before obtained multistep solutions are considered as starting estimates due to the nature of multistep collocation schemes.

    In this regard, the most recent article [7] is state-of-the-art where multistep collocation schemes have been used to solve a second-kind Fredholm integral equation, and most significantly, the strategy of employing graded meshes has been proposed to tackle weakly singular FIEs.

    Studies of [13,14,15] are elaborate evidences for efficiency as well as effectiveness of performing multistep collocation techniques encountering FIEs [7], VIEs [13,14,15], and DVIEs[9,10,11,12], within differential and/or integral equations together with prominent delay arguments. Since these classes of methods provide threefold benefits in terms of easy implementation, high-order convergent solutions, as well as the lowest degrees of freedom and complexity. Conte et al. [14] utilized a multistep collocation strategy to tackle VIEs of second-kind, which may be considered an introduction to multistep collocation methods.

    The outline of this study is as follows: The foundation of Section 2 relies on the construction of multistep collocation techniques for numerically solving second-kind VIEs along with vanishing delays on uniform meshes. A comprehensive investigation of collocation-based multistep methods in terms of existence and uniqueness of multistep collocation approximations, error analysis, and more precisely convergence results has been gathered in Section 3. Section 4 presents a numerical illustration of the performance of the proposed numerical scheme. In addition, it will validate the attained analytical findings of Section 3. The final section stands for some concluding remarks that are concerned with VIEs of pantograph type.

    In this position, we will first prove analytical solution (exact solution) existence, uniqueness, and regularity, or equivalently solvability of pantograph type second-kind VIE (1.1). Also, the same solvability discussion for the R-step collocation solution will be given further.

    Theorem 2.1. Following regularity assumptions for under consideration vanishing delay VIE (1.1), with some k0 and J:=[0,T],

    B1Ck(D1),B2Ck(D2),gCk(J),

    are held. Then, for any given data function g and delay parameter q(0,1), DVIE (1.1) has a unique solution xCk(J).

    Proof. It completely follows from [2], most precisely proof of Theorem 2.3.11.

    As we know, multistep collocation approximations for vanishing delay VIE (1.1) are given in the following manner

    u(Tm+τh)=R1r=0Ψr(τ)xmr+Ms=1ρs(τ)Xm,s,τ[0,1],mR1, (2.1)

    which lie in S(1)M+R1(δh), a discontinuous piecewise polynomial space being extensively used for collocation-based approaches, which is defined like this

    S(1)μ1(δh):={uL2(J):u|σm:=(Tm,Tm+1]πμ1,s.t.m=0,,N1},

    where πμ1 stands for real polynomials having degrees not greater than μ1. In addition, δh is considered as the uniform mesh of the introduced interval J and is like the following

    δh:={Tm:=mh;s.t.m:=0,,N;h:=T/N}.

    A solution approximation (2.1) is given with the below considerations

    Ψr(τ)=Mp=1(τcp)(rcp)R1p=0,pr(τ+p)(r+p),ρs(τ)=R1p=0(τ+p)(cs+p)Mp=1,ps(τcp)(cscp). (2.2)

    In addition, we have assumed that Xm,s=u(Tm,s) together with xmr=u(Tmr).

    Furthermore, Zh stands for collocation points as bellow

    Zh:={Tm,p:=Tm+cphs.t.0<c1<<cM1;0mN1}.

    Subsequently, the multistep collocation equation of (1.1), is obtained for p:=1,,M and m:=R,,N1 at the collocation points of Tm,p as what comes

    Xm,p=g(Tm,p)+Tm,p0B1(Tm,p,τ)um(τ)dτ+qTm,p0B2(Tm,p,τ)um(τ)dτ. (2.3)

    Next, putting the multistep collocation polynomial (2.1) into Eq (2.3), gets us to the following multistep collocation system for m:=R1,,N1

    Xm,p=g(Tm,p)+hR2i=010B1(Tm,p,Ti+τh)u(Ti+τh)dτ+hm1i=R110B1(Tm,p,Ti+τh)(R1r=0Ψr(τ)xir+Ms=1ρs(τ)Xi,s)dτ+hcp0B1(Tm,p,Tm+τh)(R1r=0Ψr(τ)xmr+Ms=1ρs(τ)Xm,s)dτ+hR2i=010B2(Tm,p,Ti+τh)u(Ti+τh)dτ+hqm,p1i=R110B2(Tm,p,Ti+τh)(R1r=0Ψr(τ)xir+Ms=1ρs(τ)Xi,s)dτ+hηm,p0B2(Tm,p,Tqm,p+τh)(R1r=0Ψr(τ)xqm,pr+Ms=1ρs(τ)Xqm,p,s)dτ. (2.4)

    It is worthy to note that, according to the key monograph by Brunner [16], tackling with vanishing delay VIEs is a bit sophisticated in comparison to VIEs without any vanishing delay term or even to a non-vanishing one. In more details, let define qTm,p like what follows:

    qTm,p:=Tqm,p+ηm,ph,

    in which taking qm,p equal and/or less than m, along with ηm,p[0,1) yields to have delay term of qTm,p being located in the interval [Tqm,p,Tqm,p+1].

    Thereafter, determination of the location of the variable delay term, i.e., qTm,p in the interval of [Tqm,p,Tqm,p+1] leads to encounter the threefold phases, which come in the subsequent.

    Phase I. qTm,1>Tm, equivalently qm,1=m.

    Phase II. qTm,1Tm<qTm,M, it means that if there is an integer, namely ii, such that for 0<ii<M, then we have qm,ii=m1, together with qm,ii+1=m.

    Phase III. qTm,MTm, which indicates that for integers like qm and ii, such that 0<qm<m1 and 0<ii<M, accordingly qm,ii=qm together with qm,ii+1=qm+1 are satisfied.

    Therefore, the phase I multistep collocation equation at Tm,p has the following form:

    Xm,p=g(Tm,p)+hR2i=010B1(Tm,p,Ti+h)u(Ti+τh)dτ+hm1i=R110B1(Tm,p,Ti+τh)(R1r=0Ψr(τ)xir+Ms=1ρs(τ)Xi,s)dτ+hcp0B1(Tm,p,Tm+τh)(R1r=0Ψr(τ)xmr+Ms=1ρs(τ)Xm,s)dτ+hR2i=010B2(Tm,p,Ti+τh)u(Ti+τh)dτ+hm1i=R110B2(Tm,p,Ti+τh)(R1r=0Ψr(τ)xir+Ms=1ρs(τ)Xi,s)dτ+hηm,p0B2(Tm,p,Tm+τh)(R1r=0Ψr(τ)xmr+Ms=1ρs(τ)Xm,s)dτ. (2.5)

    Then, with respect to the below notations, which are associated with indices of i:=0,,m; p:=1,,M; r:=0,,R1; and s:=1,,M,

    x(1)i:=(xiR+1,xiR+2,,xi)T;Xm:=(Xm,1,Xm,2,,Xm,M)T,
    Gm:=(g(Tm,1),g(Tm,2),,g(Tm,M))T,
    ˉBm,i1:=(10B1(Tm,p,Ti+τh)Ψr(τ)dτ)p,r,ˆBm,i1:=(10B1(Tm,p,Ti+τh)ρs(τ)dτ)p,s,
    ˉBm,i2:=(10B2(Tm,p,Ti+τh)Ψr(τ)dτ)p,r,ˆBm,i2:=(10B2(Tm,p,Ti+τh)ρs(τ)dτ)p,s,
    Λi1:=(10B1(Tm,p,Ti+τh)u(Ti+τh)dτ)p,1,Λi2:=(10B2(Tm,p,Ti+τh)u(Ti+τh)dτ)p,1,
    ˉBm1:=(cp0B1(Tm,p,Tm+τh)Ψr(τ)dτ)p,r,ˆBm1:=(cp0B1(Tm,p,Tm+τh)ρs(τ)dτ)p,s;
    ˉBm2:=(ηm,p0B2(Tm,p,Tm+τh)Ψr(τ)dτ)p,r,ˆBm2:=(ηm,p0B2(Tm,p,Tm+τh)ρs(τ)dτ)p,s.

    Equation (2.5) briefly becomes

    (Ih(ˆBm1+ˆBm2))Xm=Gm+hR2i=0(Λi1+Λi2)+h(ˉBm1+ˉBm2)x(1)m+hm1i=R1((ˉBm,i1+ˉBm,i2)x(1)i+(ˆBm,i1+ˆBm,i2)Xi). (2.6)

    Phase II consists of two cases; more directly, for ii<M, we have firstly, qm,ii=m1, second qm,ii+1=m. Case I includes iii such that (2.4) takes form of below

    Xm,p=g(Tm,p)+hR2i=010B1(Tm,p,Ti+τh)u(Ti+τh)dτ+hm1i=R110B1(Tm,p,Ti+τh)(R1r=0Ψr(τ)xir+Ms=1ρs(τ)Xi,s)dτ+hcp0B1(Tm,p,Tm+τh)(R1r=0Ψr(τ)xmr+Ms=1ρs(τ)Xm,s)dτ+hR2i=010B2(Tm,p,Ti+τh)u(Ti+τh)dτ+hm2i=R110B2(Tm,p,Ti+τh)(R1r=0Ψr(τ)xir+Ms=1ρs(τ)Xi,s)dτ+hηm,p0B2(Tm,p,Tm1+τh)(R1r=0Ψr(τ)xm1r+Ms=1ρs(τ)Xm1,s)dτ. (2.7)

    In addition, for p>ii, the multistep collocation equation at Tm,p is

    Xm,p=g(Tm,p)+hR2i=010B1(Tm,p,Ti+τh)u(Ti+τh)dτ+hm1i=R110B1(Tm,p,Ti+τh)(R1r=0Ψr(τ)xir+Ms=1ρs(τ)Xi,s)dτ+hcp0B1(Tm,p,Tm+τh)(R1r=0Ψr(τ)xmr+Ms=1ρs(τ)Xm,s)dτ+hR2i=010B2(Tm,p,Ti+τh)u(Ti+τh)dτ+hm1i=R110B2(Tm,p,Ti+τh)(R1r=0Ψr(τ)xir+Ms=1ρs(τ)Xi,s)dτ+hηm,p0B2(Tm,p,Tm+τh)(R1r=0Ψr(τ)xmr+Ms=1ρs(τ)Xm,s)dτ. (2.8)

    Phase II multistep collocation equation is presented by

    (Ih(ˆBm1+ˆˆBm2))Xm=Gm+hR2i=0(Λi1+Λi2)+h(ˉBm1+ˉˉBm2)x(1)m+hm1i=R1((ˉBm,i1+ˉBm,i2)x(1)i+(ˆBm,i1+ˆBm,i2)Xi), (2.9)

    where

    ˉˉBm2:=diag(0,,0,1,,1)ˉBm2;ˆˆBm2:=diag(0,,0,1,,1)ˆBm2,

    such that in both ii is the number of zeros. We are now concerned with phase III. Herein, we face two cases, such that for integers in terms of qm and ii with assuming 0<qm<m together with 0<ii<M, we then have qm,ii=qm1 and qm,ii+1=qm. Consequently, the phase III multistep collocation equation is as follows:

    Xm,p=g(Tm,p)+hR2i=010B1(tm,p,Ti+τh)u(Ti+τh)dτ+hm1i=R110B1(Tm,p,Ti+τh)(R1r=0Ψr(τ)xir+Ms=1ρs(τ)Xi,s)dτ+hcp0B1(Tm,p,Tm+τh)(R1r=0Ψr(τ)xmr+Ms=1ρs(τ)Xm,s)dτ+hR2i=010B2(Tm,p,Ti+τh)u(Ti+τh)dτ+hqm1i=R110B2(Tm,p,Ti+τh)(R1r=0Ψr(τ)xir+Ms=1ρs(τ)Xi,s)dτ+hηm,p0B2(Tm,p,Tqm+τh)(R1r=0Ψr(τ)xqmr+Ms=1ρs(τ)Xqm,s)dτ, (2.10)

    which can be written

    (IhˆBm1)Xm=Gm+hR2i=0(Λi1+Λi2)+hˉBm1x(1)m+hm1i=R1(ˉBm,i1x(1)i+ˆBm,i1Xi)+hqm1i=R1(ˉBm,i2x(1)i+ˆBm,i2Xi)+h(ˉˉBqm2x(1)qm+ˆˆBqm2Xqm), (2.11)

    in which

    ˉˉBqm2:=diag(0,,0,1,,1)ˉBm,qm2,ˆˆBqm2:=diag(0,,0,1,,1)ˆBm,qm2,

    are satisfied such that the number of zeros is ii.

    Supposing

    ˆBm1B1β,ˆBm2B2β, (2.12)

    where

    β:=Ms=1τ[0,1]sup (2.13)

    may lead to helpful underlying bounds, which will be used in solvability proofs. Then, we have what follows:

    \begin{equation} \Vert\hat{\hat{B}}_{2}^{m}\Vert_{\infty}\leqslant\Vert \hat{B}_{2}^{m}\Vert_{\infty}\leqslant \Vert {B}_{2}\Vert_{\infty}\beta . \end{equation} (2.14)

    In the subsequent discussion, we will investigate the solvability of the system derived from the numerical technique. Hence, by means of the underlying theorem, we have demonstrated the most principle idea of compact operators in the case of second-kind vanishing delay VIEs.

    Theorem 2.2. Recall \mathbb{X}: = {C}^{M+R}([0, \mathfrak{T}]) . Let linear and continuous operators \mathfrak{B}_{1}, \mathfrak{B}_{2}:\mathbb{X}\rightarrow S^{(-1)}_{M+R-1}(\delta _{h}) as follows:

    (\mathfrak{B}_{1}x)(T): = T\int_{0}^{\mathfrak{T}}{B}_{1}(T,\tau (T,\nu))x(\tau (T,\nu)){d}\nu,
    (\mathfrak{B}_{2}x)(T): = (qT)\int_{0}^{\mathfrak{T}}{B}_{2}(T,q\tau (T,\nu))x(q\tau (T,\nu))\mathrm{d}\nu.

    Then, for operator format of (1.1), more precisely the following

    \Big(\mathcal{I}-(\mathfrak{B}_{1}+\mathfrak{B}_{2})\Big)x = g,

    \mathfrak{B}_{1} and \mathfrak{B}_{2} stand for compact operators.

    Proof. First, we transform intervals of non-delay and delay integrations, respectively [0, T] and [0, qT] into [0, \mathfrak{T}] with the constant \mathfrak{T} , through a transformation like

    \begin{align} &\tau :[0,\mathfrak{T}]\times[0,\mathfrak{T}]\longrightarrow[0,\mathfrak{T}]\\ &\tau (T,\nu): = T\nu . \end{align} (2.15)

    In this way, under study equation of (1.1) may be transformed into its operator form as

    \Big(\mathcal{I}-(\mathfrak{B}_{1}+\mathfrak{B}_{2})\Big)x = g.

    Suppose

    \mathbb{B}_{1}: = \lbrace \mathfrak{B}_{1}x:x \in {C}^{M+R}([0,\mathfrak{T}])\rbrace ,\quad \mathbb{B}_{2}: = \lbrace \mathfrak{B}_{2}x:x \in {C}^{M+R}([0,\mathfrak{T}])\rbrace .

    Based on [17], we aim to prove that \mathbb{B}_{1} and \mathbb{B}_{2} are equicontinuous and uniformly bounded, and then, immediately, thanks to Arzela-Ascoli Theorem, the compactness of operators \mathfrak{B}_{1} and \mathfrak{B}_{2} can be concluded. By assuming {L}_{1}: = \Vert {B}_{1}\Vert _{\infty} = \sup \vert {B}_{1}(T, \tau(T, \nu))\vert and {L}_{2}: = \Vert {B}_{2}\Vert _{\infty} = \sup \vert {B}_{2}(T, \tau(T, \nu))\vert for T, \tau \in [0, \mathfrak{T}] , it is straightforward to see

    \begin{align*} &\Vert \mathfrak{B}_{1} x\Vert_{\infty} = \underset{T\in [0,\mathfrak{T}]}{\sup}\vert \mathfrak{B}_{1} x(T)\vert\\ &\; \; \; \; \; \; \; \; \; \; \; \; = \underset{T\in [0,\mathfrak{T}]}{\sup}\vert T\int_{0}^{\mathfrak{T}}{B}_{1}(T,\tau (T,\nu))x(\tau (T,\nu)){d}\nu\vert \leq {\mathfrak{T}}^{2}{L}_{1}\Vert x\Vert _{\infty} < \infty. \end{align*}

    In a similar way,

    \Vert \mathfrak{B}_{2} x\Vert_{\infty}\leq q {\mathfrak{T}}^{2}{L}_{2}\Vert x\Vert _{\infty} < \infty .

    Thereby, the uniformly boundedness property of sets \mathbb{B}_{1} and \mathbb{B}_{2} has been derived. When it comes to continuity, we only prove that \mathfrak{B}_{2} is a continuous operator; the proof for \mathfrak{B}_{1} is in the same manner. Due to the assumption that kernel {B}_{2} and solution y are sufficiently smooth on their domains, for all T, \bar{T} from [0, \mathfrak{T}]

    \begin{equation*} \lim\limits_{T \to \bar{T}} \vert x(\tau(\bar{T},\nu))-x(\tau(T,\nu))\vert = 0, \end{equation*}

    and

    \begin{equation*} \lim\limits_{T \to \bar{T}} \vert {B}_{2}(\bar{T},q\tau(\bar{T},\nu))-{B}_{2}(T,q\tau(T,\nu))\vert = 0, \end{equation*}

    are satisfied. Hence, we conduct the following when T\longrightarrow \bar{T} .

    \begin{align*} &\vert \mathfrak{B}_{2} x(\bar{T})-\mathfrak{B}_{2}x(T)\vert\\ &\; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; = \vert\int_{0}^{\mathfrak{T}}\Big({B}_{2}(\bar{T},q\tau(\bar{T},\nu))x(q\tau(\bar{T},\nu))-{B}_{2}(T,q\tau(T,\nu))x(q\tau(T,\nu))\Big)\mathrm{d}\nu\vert \\ &\; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; = \vert \int_{0}^{\mathfrak{T}}\Big({B}_{2}(\bar{T},q\tau(\bar{T},\nu))x(q\tau(\bar{T},\nu))-{B}_{2}(\bar{T},q\tau(\bar{T},\nu))x(q\tau(T,\nu))\\ &\; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; +{B}_{2}(\bar{T},q\tau(\bar{T},\nu))x(q\tau(T,\nu))-{B}_{2}(T,q\tau(T,\nu))x(q\tau(T,\nu))\Big)\mathrm{d}\nu\vert \\ &\; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \leq \int_{0}^{\mathfrak{T}} \vert {B}_{2}(\bar{T},q\tau(\bar{T},\nu))[x(q\tau(\bar{T},\nu))-x(q\tau(T,\nu))]\vert \mathrm{d}\nu \\ &\; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; +\int_{0}^{\mathfrak{T}}\vert[{B}_{2}(\bar{T},q\tau(\bar{T},\nu))-{B}_{2}(T,q\tau(T,\nu))]x(q\tau(T,\nu))\vert \mathrm{d}\nu \\ &\; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \leq \mathfrak{T}\Big(\Vert{B}_{2}\Vert_{\infty}\sup \vert x(q\tau(\bar{T},\nu))-x(q\tau(T,\nu))\vert \\ &\; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \; +\sup \vert{B}_{2}(\bar{T},q\tau(\bar{T},\nu))-{B}_{2}(T,q\tau(T,\nu))\vert \Vert x\Vert_{\infty}\Big)\longrightarrow 0. \end{align*}

    This fulfills the proof.

    The upcoming theorem, i.e., solvability of numerical solutions, has strong ties with the compactness concept of second-kind pantograph delay VIEs operators. Subsequent theorem demonstrates solvability for numerically obtained multistep collocation solution belonging to (2.3).

    Theorem 2.3. Consider {B}_{1}(T, \tau)\in {C}^{M+R}({D}_{1}) , {B}_{2}(T, \tau)\in {C}^{M+R}({D}_{2}) , and g \in {C}^{M+R}(J) for vanishing DVIE (1.1). Then, multistep collocation equation (2.3) is uniquely solvable for a unique numerical solution {u}\in S^{(-1)}_{M+R-1}(\delta _{h}) , which has been provided by relation (2.1) in a general subinterval {\sigma}_{m}: = ({T}_{m}, {T}_{m+1}] of the uniform \delta _{h} mesh, such that for a given \mathfrak{h} > 0 we have

    \begin{equation*} 0 < h < \mathfrak{h}: = \frac{1}{M(\Vert {B}_{1}\Vert_{\infty} +\Vert {B}_{2}\Vert_{\infty})\beta}, \end{equation*}

    where \beta is defined in (2.13) and M is the number of collocation parameters.

    Proof. Overall, we aim to show that coefficient matrices in phases I, II, and III, respectively, in terms of

    \begin{align} &I-h(\hat{B}_{1}^{m}+\hat{B}_{2}^{m}),\\ &I-h(\hat{B}_{1}^{m}+\hat{\hat{B}}_{2}^{m}),\\ &I-h\hat{B}_{1}^{m}, \end{align} (2.16)

    with 0\leq m\leq N-1 are invertible. For this sake, take into account the obtained multistep collocation equations in three phases, mainly (2.6), (2.9), and (2.11). Recalling achieved bounds of (2.12) and (2.14), there is a well-defined

    \begin{equation*} \mathfrak{h}: = \frac{1}{M(\Vert {B}_{1}\Vert_{\infty}+\Vert {B}_{2}\Vert_{\infty})\beta} > 0, \end{equation*}

    such that in a generic subinterval {\sigma}_{m}: = ({T}_{m}, {T}_{m+1}] of \delta_{h} with 0 < h < \mathfrak{h} for 0\leq m\leq N-1 , we have the followings

    \begin{align} &\Vert h(\hat{B}_{1}^{m}+\hat{B}_{2}^{m})\Vert_{\infty} < 1,\\ &\Vert h(\hat{B}_{1}^{m}+\hat{\hat{B}}_{2}^{m})\Vert_{\infty} < 1,\\ &\Vert h\hat{B}_{1}^{m}\Vert_{\infty} < 1. \end{align} (2.17)

    Thereafter, thanks to the Neumann theorem, we can display the desired assertion as follows:

    \begin{align} &\Vert \Big(I-h(\hat{B}_{1}^{m}+\hat{B}_{2}^{m})\Big)^{-1}\Vert_{\infty} < \infty,\\ &\Vert \Big(I-h(\hat{B}_{1}^{m}+\hat{\hat{B}}_{2}^{m})\Big)^{-1}\Vert_{\infty} < \infty,\\ &\Vert \Big(I-h\hat{B}_{1}^{m}\Big)^{-1}\Vert_{\infty} < \infty , \end{align} (2.18)

    which asserts that multistep collocation equation (2.3) is uniquely solvable in all three phases for obtaining a unique approximation of (2.1).

    In this position, we are interested in concerns with convergence analysis. In view of the operator form of the under study second-kind pantograph DVIE (1.1), specifically

    \begin{equation} \Big(\mathcal{I}-(\mathfrak{B}_{1}+\mathfrak{B}_{2})\Big)x = g, \end{equation} (2.19)

    with respect to compact operators \mathfrak{B}_{1} and \mathfrak{B}_{2} introduced in Theorem 2.2, the following properties equip us to establish the convergence theorem in the sequel. In order to be convenient, we name \mathfrak{B}_{1}+\mathfrak{B}_{2} with \mathscr{B} , which is a compact operator, as Theorem 2.2 shows. The idea of using a projection interpolation operator for attaining convergence is an immediate outcome of compactness. Regarding this, projection interpolation operator like \mathscr{P} is defined in the following for an arbitrary function of x ,

    \begin{align} &\mathscr{P}:C(J: = [0,\mathfrak{T}])\longrightarrow S^{(-1)}_{M+R-1}(\delta_{h}),\\ &(\mathscr{P}x)({T}_{m,p}) = x({T}_{m,p}),\qquad {T}_{m,p}\in {\mathbb{Z}}_{h}. \end{align} (2.20)

    Afterwards, in order to obtain operator form of error equation, influence equation (1.1), and multistep collocation equation (2.3) by projection operator of \mathscr{P} as follows

    \begin{align} &\mathscr{P}x = \mathscr{P}\mathscr{B}x+\mathscr{P}g,\\ &\mathscr{P}{u} = \mathscr{P}\mathscr{B}{u}+\mathscr{P}g. \end{align} (2.21)

    The operator error equation is obtained straightforwardly in the following way:

    \begin{align} &e = \mathscr{P}e\\ &\; = \mathscr{P}x-{u}\\ &\; = \mathscr{P}\mathscr{B}x-\mathscr{P}\mathscr{B}{u}\\ &\; = \mathscr{P}\mathscr{B}(\mathscr{P}e)+\mathscr{P}\mathscr{B}(x-\mathscr{P}x)\\ &\; = \mathscr{P}\mathscr{B}e+\mathscr{P}\mathscr{B}(\mathcal{I}-\mathscr{P})x\\ &\; = \mathscr{P}\mathscr{B}e+\mathcal{K} , \end{align} (2.22)

    in which we have assumed \mathcal{K}: = \mathscr{P}\mathscr{B}(\mathcal{I}-\mathscr{P})x . Additionally, for the sake of simplicity, take \mathscr{W}: = \mathscr{P}\mathscr{B} . From (2.22), it can easily be seen that

    \begin{equation} \Vert e\Vert_{\infty}\leq \Vert(\mathcal{I}-\mathscr{W})^{-1}\Vert_{\infty}\Vert \mathcal{K}\Vert_{\infty}. \end{equation} (2.23)

    Lemma 2.1. By supposing {B}_{1}\in {C}^{M+R}({D}_{1}) , {B}_{2}\in {C}^{M+R}({D}_{2}) , and g \in{C}^{M+R}(J) , then the following key results of

    \begin{equation} \lim\limits_ {h \to 0} \Vert \mathscr{W}-\mathscr{B}\Vert_{\infty} = 0, \end{equation} (2.24)
    \begin{equation} \Vert (\mathcal{I}-\mathscr{W})^{-1}\Vert_{\infty} < \infty \end{equation} (2.25)

    are held.

    Proof. On the one hand \mathscr{W}: = \mathscr{P}\mathscr{B} , on the other hand, according to Theorem 2.2, \mathscr{B} \in \mathcal{L}(S^{(-1)}_{M+R-1}(\delta_{h})) is a compact operator, thus it can be immediately concluded that \mathscr{W} is uniformly convergent to \mathscr{B} , hence (2.24) is asserted. Again, due to the compactness of \mathscr{B} , based on Theorem 2.6 in the study [18], (\mathcal{I}-\mathscr{B}) is an invertible operator; however, its inverse is bounded. Most specifically,

    \begin{equation} \Vert {(\mathcal{I}-\mathscr{B})}^{-1}\Vert_{\infty} < \infty . \end{equation} (2.26)

    Regarding well-established (2.24) and (2.26) yields to

    \begin{equation} \Vert \mathscr{W}-\mathscr{B}\Vert_{\infty}\Vert {(\mathcal{I}-\mathscr{B})}^{-1}\Vert_{\infty}\leq 1. \end{equation} (2.27)

    (2.27) results in boundedness of operator \mathscr{W} , thereby, thanks to the Neumann Theorem, (\mathcal{I}-\mathscr{W}) is invertible and bounded simultaneously. In order to attain a bound of \Vert {(\mathcal{I}-\mathscr{W})}^{-1}\Vert_{\infty} we pursue what follows. Operator multistep collocation equation of (2.21) can be written for suitably assumed \psi \in S^{(-1)}_{M+R-1}(\delta_{h}) such that

    \begin{equation} {u} = \mathscr{W}{u}+\psi , \end{equation} (2.28)

    in which most importantly, \Vert \psi\Vert_{\infty} = 1 has been taken. First, subtracting \mathscr{B}{u} from (2.28), then multiplying by {(\mathcal{I}-\mathscr{B})}^{-1} derives the following relation:

    \begin{equation} {u} = {(\mathcal{I}-\mathscr{B})}^{-1}(\mathscr{W}-\mathscr{B}){u}+{(\mathcal{I}-\mathscr{B})}^{-1}\psi . \end{equation} (2.29)

    Taking norm of (2.29), with respect to (2.27) yields

    \begin{align} &\Vert {u}\Vert _{\infty} \leq \frac{1}{1-\Vert (\mathcal{I}-\mathscr{B})^{-1}(\mathscr{W}-\mathscr{B})\Vert _{\infty}}\Vert (\mathcal{I}-\mathscr{B})^{-1}\psi\Vert_{\infty}\\ &\; \; \; \; \; \; \; \leq 2 \Vert (\mathcal{I}-\mathscr{B})^{-1}\Vert < \infty . \end{align}

    This ends the proof with obtaining the desired result of \Vert {(\mathcal{I}-\mathscr{W})}^{-1}\Vert_{\infty} = \Vert {u}\Vert _{\infty} < \infty .

    We will just exploit the spectral radius concept of the matrix based on collocation parameters in the difference equation arising from the method's error in concluding convergence results. We now give the convergence analysis in a theorem frame.

    Theorem 3.1. In pantograph DVIE (1.1), we have supposed that g\in {C}^{M+R}(J) , {B}_{1}\in {C}^{M+R}({D}_{1}) , and {B}_{2}\in {C}^{M+R}({D}_{2}) . For \varepsilon: = x-u and P: = M+R , suppose that there is a start error of \Vert \varepsilon\Vert_{\infty}: = \underset{t\in[0, {t}_{r-1}]}{\sup}\mid \varepsilon(t)\mid = \mathcal{O}(h^{P}) . Also, with \rho , which denotes the spectral radius of the matrix \mathbf{S} , \rho(\mathbf{S}) < 1 is satisfied. It should be noted that \mathbf{S} will be defined further in the proof stream. Then, the uniquely attained multistep collocation solution {u} uniformly converges to the exact solution x , holding the convergence order of M+R

    \begin{equation*} \Vert x-{u}\Vert_{\infty}: = \underset{t\in J: = [0,T]}{\sup}\mid x(t)-{u}(t)\mid \leq c{h}^{P}. \end{equation*}

    Proof. For the exact solution x with any \tau \in [0, 1] , the following representation is satisfied by considering the Peano's theorem on the representation of interpolation error given in the Corollary 1.8.2 of [16].

    \begin{equation} x({T}_{m}+\tau h) = \sum\limits_{r = 0}^{R-1}\psi _{r}(\tau)x({T}_{m-r})+\sum\limits_{p = 1}^{M}\rho _{p}(\tau)x({T}_{m,r}) +{h}^{P}{\tilde{R}}_{M,R,m}(\tau), \end{equation} (3.1)

    where the bases \psi _{r}(\tau) and \rho _{p}(\tau) have been previously defined. Also, the remainder function is given as below

    \begin{equation*} {\tilde{R}}_{M,R,m}(\tau): = \int_{-R+1}^{1}{\tilde{L}}_{M,R}(\tau,v){x}^{(p)}({T}_{m}+vh)\mathrm{d}v, \end{equation*}

    in which

    \begin{align*} &{\tilde{L}}_{M,R}(\tau,v): = \\ &\frac{1}{(M+R-1)!}\Big((\tau-v)_{+}^{M+R-1}-\sum\limits_{r = 0}^{R-1}\psi _{r}(\tau)(-k-v)_{+}^{M+R-1}-\sum\limits_{p = 1}^{M}\rho _{p}(\tau)({c}_{p}-v)_{+}^{M+R-1}\Big). \end{align*}

    Similarly, we have the error representation for \tau \in [0, 1] as follows:

    \begin{equation} \varepsilon({T}_{m}+\tau h) = \sum\limits_{r = 0}^{R-1}\psi _{r}(\tau)\varepsilon _{m-r}+\sum\limits_{p = 1}^{M}\rho _{p}(\tau)\varepsilon _{m,p}+h^{P}{\tilde{R}}_{M,R,m}(\tau), \end{equation} (3.2)

    such that \varepsilon _{m-r}: = x(t_{m-r})-{u}(t_{m-r}) , and \varepsilon _{m, p}: = x(t_{m, p})-{u}(t_{m, p}) , are held. Recalling the multistep collocation system of (2.4) and considering the fact that the errors satisfy in (2.4), we reach to

    \begin{align} &{\varepsilon}({T}_{m,p}) = h\int_{0}^{{c}_{p}}{B}_{1}({T}_{m,p},{T}_{m}+\tau h){\varepsilon}({T}_{m}+\tau h)\mathrm{d}\tau\\ &\; \; \; \; \; \; \; \; \; \; \; \; +h\sum\limits_{i = 0}^{m-1}\int_{0}^{1}{B}_{1}({T}_{m,p},{T}_{i}+\tau h){\varepsilon}({T}_{i}+\tau h)\mathrm{d}\tau\\ &\; \; \; \; \; \; \; \; \; \; \; \; +h\int_{0}^{{\eta}_{m,p}}{B}_{2}({T}_{m,p},{T}_{{q}_{m,p}}+\tau h){\varepsilon}({T}_{{q}_{m,p}}+\tau h)\mathrm{d}\tau\\ &\; \; \; \; \; \; \; \; \; \; \; \; +h\sum\limits_{i = 0}^{{q}_{m,p}-1}\int_{0}^{1}{B}_{2}({T}_{m,p},{T}_{i}+\tau h){\varepsilon}({T}_{i}+\tau h)\mathrm{d}\tau . \end{align} (3.3)

    Subsequently, by taking advantage of (3.2) and assuming that in phase I we have {q}_{m, p} = m , it is straightforward to get to the following, which are indeed the second rows of our error matrices.

    \begin{align} &{\varepsilon}_{m,p} = h\sum\limits_{r = 0}^{R-1}(\int_{0}^{{c}_{p}}{B}_{1}({T}_{m,p},{T}_{m}+\tau h){\psi}_{r}(\tau)\mathrm{d}\tau)\varepsilon_{m-r}\\ &\; \; \; \; \; \; \; \; \; +h\sum\limits_{p = 1}^{M}(\int_{0}^{{c}_{p}}{B}_{1}({T}_{m,p},{T}_{m}+\tau h){\rho}_{p}(\tau)\mathrm{d}\tau)\varepsilon_{m,p}\\ &\; \; \; \; \; \; \; \; \; +h\sum\limits_{i = 0}^{m-1}\sum\limits_{r = 0}^{R-1}(\int_{0}^{1}{B}_{1}({T}_{m,p},{T}_{i}+\tau h){\psi}_{r}(\tau)\mathrm{d}\tau)\varepsilon_{i-r} \\ &\; \; \; \; \; \; \; \; \; +h\sum\limits_{i = 0}^{m-1}\sum\limits_{p = 1}^{M}(\int_{0}^{1}{B}_{1}({T}_{m,p},{T}_{i}+\tau h){\rho}_{p}(\tau)\mathrm{d}\tau)\varepsilon_{i,p}\\ &\; \; \; \; \; \; \; \; \; +h\sum\limits_{r = 0}^{R-1}(\int_{0}^{{\eta}_{m,p}}{B}_{2}({T}_{m,p},{T}_{m}+\tau h){\psi}_{r}(\tau)\mathrm{d}\tau)\varepsilon_{m-r} \\ &\; \; \; \; \; \; \; \; \; +h\sum\limits_{p = 1}^{M}(\int_{0}^{{\eta}_{m,p}}{B}_{2}({T}_{m,p},{T}_{m}+\tau h){\rho}_{p}(\tau)\mathrm{d}\tau)\varepsilon_{m,p}\\ &\; \; \; \; \; \; \; \; \; +h\sum\limits_{i = 0}^{m-1}\sum\limits_{r = 0}^{R-1}(\int_{0}^{1}{B}_{2}({T}_{m,p},{T}_{i}+\tau h){\psi}_{r}(\tau)\mathrm{d}\tau)\varepsilon_{i-r}\\ &\; \; \; \; \; \; \; \; \; +h\sum\limits_{i = 0}^{m-1}\sum\limits_{p = 1}^{M}(\int_{0}^{1}{B}_{2}({T}_{m,p},{T}_{i}+\tau h){\rho}_{p}(\tau)\mathrm{d}\tau)\varepsilon_{i,p}+\mathcal{O}(h^P) . \end{align} (3.4)

    Now, it requires to assume what follows

    {E}^{(1)}_{m}: = [\varepsilon _{m-R+1},\varepsilon _{m-R+2},\ldots,\varepsilon _{m}]^{T},\qquad {E}^{(2)}_{m}: = [\varepsilon _{m,1},\varepsilon _{m,2},\ldots,\varepsilon _{m,M}]^{T}.

    Additionally, by means of Taylor's expansion, it is straightforward to define the following:

    \begin{align} &\int_{0}^{1}{B}_{1}({T}_{m,p},{T}_{i}+\tau h){\psi}_{r}(\tau)\mathrm{d}\tau = {B}_{1}({T}_{m},{T}_{m})\sigma_{r}+\mathcal{O}(h),\\ &\int_{0}^{1}{B}_{1}({T}_{m,p},{T}_{i}+\tau h){\rho}_{p}(\tau)\mathrm{d}\tau = {B}_{1}({T}_{m},{T}_{m})\gamma_{p}+\mathcal{O}(h),\\ &\int_{0}^{{c}_{p}}{B}_{1}({T}_{m,p},{T}_{m}+\tau h){\psi}_{r}(\tau)\mathrm{d}\tau = {B}_{1}({T}_{m},{T}_{m}){l}_{pr}+\mathcal{O}(h),\\ &\int_{0}^{{c}_{p}}{B}_{1}({T}_{m,p},{T}_{m}+\tau h){\rho}_{r}(\tau)\mathrm{d}\tau = {B}_{1}({T}_{m},{T}_{m}){f}_{pr}+\mathcal{O}(h),\\ &\int_{0}^{{\eta}_{m,p}}{B}_{2}({T}_{m,p},{T}_{m}+\tau h){\psi}_{r}(\tau)\mathrm{d}\tau = {B}_{2}({T}_{m},{T}_{m}){\hat{l}}_{pr}+\mathcal{O}(h),\\ &\int_{0}^{{\eta}_{m,p}}{B}_{2}({T}_{m,p},{T}_{m}+\tau h){\rho}_{r}(\tau)\mathrm{d}\tau = {B}_{2}({T}_{m},{T}_{m}){\hat{f}}_{pr}+\mathcal{O}(h), \end{align} (3.5)

    where

    {\sigma}_{r}: = \underset{r = 0,\cdots,R-1}{\int_{0}^{1}\psi_{r}(\tau)\mathrm{d}\tau},\qquad{\gamma}_{p}: = \underset{p = 1,\cdots,M}{\int_{0}^{1}\rho_{p}(\tau)\mathrm{d}\tau}.

    In order to achieve error matrices, we use the idea in the article [19]. To do so, assume \tau = 1 in (3.2), which leads to

    E_{n}^{(1)} = \Omega E_{n-1}^{(1)}+\Upsilon E_{n-1}^{(2)}+h^{P}{\tilde{R}}_{M,R,m-1},

    such that {\tilde{R}}_{M, R, m-1} = \begin{bmatrix} & {\mathbf{0}}_{R-1, 1} & \\ & {\tilde{R}}_{M, R, m-1}(1) & \end{bmatrix} .

    Therefore, we have a phase I error system as follows:

    \begin{align} & \begin{bmatrix} &{\mathbf{I}}_{R,R}&{\mathbf{0}}_{R,M}&\\ &&&\\ &h(\mathbf{{B}_{1}}\mathbf{L}+\mathbf{{B}_{2}}\mathbf{\hat{L}})&\mathbf{I}-h(\mathbf{{B}_{1}}\mathbf{F}+\mathbf{{B}_{2}}\mathbf{\hat{F}})& \end{bmatrix} \begin{bmatrix} &{E}^{(1)}_{m}&\\ &\\ &{E}^{(2)}_{m}& \end{bmatrix}\\ & = \begin{bmatrix} &\Omega &\Upsilon &\\ &&&\\ &(\mathbf{{B}_{1}}+\mathbf{{B}_{2}})\mathbf{\Sigma}&(\mathbf{{B}_{1}}+\mathbf{{B}_{2}})\mathbf{\Gamma}& \end{bmatrix}\\ & \begin{bmatrix} &{E}^{(1)}_{m-1}&\\ &\\ &{E}^{(2)}_{m-1}& \end{bmatrix} +\sum\limits_{i = r-1}^{m-2} \begin{bmatrix} &{\mathbf{0}}_{R,R}&{\mathbf{0}}_{R,M}&\\ &&&\\ &(\mathbf{{B}_{1}}+\mathbf{{B}_{2}})\mathbf{\Sigma}&(\mathbf{{B}_{1}}+\mathbf{{B}_{2}})\mathbf{\Sigma}& \end{bmatrix} \begin{bmatrix} &{E}^{(1)}_{i}&\\ &\\ &{E}^{(2)}_{i}& \end{bmatrix}\\ & +\mathcal{O}(h^P), \end{align} (3.6)

    where \mathbf{{B}_{1}} and \mathbf{{B}_{2}} can be readily derived and

    \mathbf{L}: = \begin{bmatrix} {l}_{pr}: = \underset{p = 1,\ldots,M,\quad r = 0,\cdots,R-1}{ \int_{0}^{{c}_{p}}\psi_{r}(\tau)\mathrm{d}\tau} \end{bmatrix},\qquad \mathbf{F}: = \begin{bmatrix} {f}_{pr}: = \underset{p = 1,\ldots,M,\quad r = 1,\cdots,M}{ \int_{0}^{{c}_{p}}\rho_{r}(\tau)\mathrm{d}\tau} \end{bmatrix},
    \mathbf{\hat{L}}: = \begin{bmatrix} {\hat{l}}_{pr}: = \underset{p = 1,\ldots,M,\quad r = 0,\ldots,R-1}{ \int_{0}^{{\eta}_{m,p}}\psi_{r}(\tau)\mathrm{d}\tau} \end{bmatrix},\qquad \mathbf{\hat{F}}: = \begin{bmatrix} {\hat{f}}_{pr}: = \underset{p = 1,\ldots,M,\quad r = 1,\ldots,M}{ \int_{0}^{{\eta}_{m,p}}\rho_{r}(\tau)\mathrm{d}\tau} \end{bmatrix},
    \mathbf{\Omega}: = \begin{bmatrix} {\mathbf{0}}_{R-1,1}&{\mathbf{I}}_{R-1}\\ \psi _{R-1}(1)&\psi _{R-2}(1),\cdots,\psi _{0}(1) \end{bmatrix},\qquad \mathbf{\Upsilon}: = \begin{bmatrix} {\mathbf{0}}_{R-1,M}\\ \rho ^{T}(1) \end{bmatrix},
    \rho (1): = [\rho _{1}(1),\rho _{2}(1),\ldots,\rho _{M}(1)]^{T}.
    \mathbf{\Sigma}: = [\sigma_{R-1},\ldots,\sigma_{0}]^{T},\qquad\mathbf{\Gamma}: = [{\gamma}_{1},\ldots,{\gamma}_{M}]^{T}.

    Then, after some manipulation, we obtain into

    \begin{equation} \begin{bmatrix} &{E}_{m}^{(1)}&\\ &\\ &{E}_{m}^{(2)}& \end{bmatrix} = \mathbf{S} \begin{bmatrix} &{E}_{m-1}^{(1)}&\\ &\\ &{E}_{m-1}^{(2)}& \end{bmatrix} +\sum\limits_{i = r-1}^{m-2} \begin{bmatrix} &{\mathbf{0}}_{R,R}&{\mathbf{0}}_{R,M}&\\ &&&\\ &\mathcal{\bar{H}}^{(i)}&\mathcal{\hat{H}}^{(i)}& \end{bmatrix} \begin{bmatrix} &{E}_{i}^{(1)}&\\ &\\ &{E}_{i}^{(2)}& \end{bmatrix} +\mathcal{O}(h^P), \end{equation} (3.7)

    in which one can derive \mathcal{\bar{H}} and \mathcal{\hat{H}} easily and the matrix of \mathbf{S} is in the following manner

    \begin{align*} &\mathbf{S}: = \left [\begin{array}{ll} &\mathbf{\Omega}\nonumber\\ &\nonumber\\ &{\Big(\mathbf{I}-h(\mathbf{{B}_{1}}\mathbf{F}+\mathbf{{B}_{2}}\mathbf{\hat{F}})\Big)}^{-1}\Big(-h\mathbf{\Omega}(\mathbf{{B}_{1}}\mathbf{L}+\mathbf{{B}_{2}}\mathbf{\hat{L}})+(\mathbf{{B}_{1}}+\mathbf{{B}_{2}})\mathbf{\Sigma}\Big) \nonumber \end{array}\right.\\ &\\ & \quad \left. \begin{array}{ll} &\mathbf{\Upsilon}\\ &\\ &{\Big(\mathbf{I}-h(\mathbf{{B}_{1}}\mathbf{F}+\mathbf{{B}_{2}}\mathbf{\hat{F}})\Big)}^{-1}\Big(-h\mathbf{\Upsilon}(\mathbf{{B}_{1}}\mathbf{L}+\mathbf{{B}_{2}}\mathbf{\hat{L}})+(\mathbf{{B}_{1}}+\mathbf{{B}_{2}})\mathbf{\Gamma}\Big)\nonumber \end{array}\right ]. \end{align*}

    Thereafter, by means of the Gronwall Lemma together with the assumption of \rho (\mathbf{S}) < 1 , we can conclude that

    \begin{equation*} \Vert {E}^{(1)}_{m}\Vert_{\infty} = \mathcal{O}({h}^{P}),\quad \Vert {E}^{(2)}_{m}\Vert_{\infty} = \mathcal{O}({h}^{P}). \end{equation*}

    Now, by taking into account (3.1), it is straightforward to see

    \begin{align*} &\Vert x-{u}\Vert_{\infty} = \sup\limits_{T\in J: = [0,\mathfrak{T}]} \mid \varepsilon (T)\mid\\ &\; \; \; \; \; \; \; \; \; \; \; \; \; \; \; \leqslant \alpha \max\limits_{0\leqslant m \leqslant N-1}\Big(\Vert {E}^{(1)}_{m}\Vert_{\infty}+\Vert {E}^{(2)}_{m}\Vert_{\infty}\Big)+\bar{\beta} {h}^{P} \leqslant C{h}^{P}, \end{align*}

    where C is a general constant along with

    \alpha: = \max \lbrace \Vert {\psi}_{r}\Vert_{\infty};r = 0,\ldots,R-1,\Vert {\rho}_{p}\Vert_{\infty};p = 1,\ldots,M,\rbrace,

    and

    \bar{\beta}: = \max \lbrace\Vert {x}^{(p)}\Vert_{\infty}, \int_{-R+1}^{0}\vert{\tilde{L}}_{M,R}(\tau,v)\vert\mathrm{d}v\rbrace,

    with \tau \in [0, 1] . Pursuing a similar process, one can deduce convergence analysis in both phase II and phase III. This is the end of the proof.

    In this present section, we concentrate on numerically testing and also verifying the previously attained convergence results of Theorem 3.1. What is more, we try various 2 -point and 3 -point multistep collocation techniques with R = 3 on uniform meshes in order to illustrate optimum {L}_{\infty} -norm convergence orders. It should be pointed out that 2 and 3 illustrate the number of collocation parameters.

    When it comes to comparing with [1], where the same second-type vanishing delay VIE has been analysed and solved, the significant distinct advantages of our proposed technique are twofold. First, the advantage of our method is that it achieves the same error results with significantly fewer spatial grid points specifically, N = 256 compared to the N = 1600 grid points used in [1]. As we know, N is a commonly used notation for displaying the number of interval divisions. Second, from simplicity and efficiency point of view, in our work a fast rate multistep strategy is implemented on uniform meshes with ease.

    Locally achieved absolute multistep collocation errors for {\sigma}_{m}: = ({T}_{m}, {T}_{m+1}] with m: = 1, \ldots, N-1 in a defined uniform {\delta}_{h} mesh are as follows:

    {e}_{m}: = \max \vert {u}({T}_{m})-{x}({T}_{m})\vert .

    Besides, rates of convergence are like the follows:

    p: = \log_{2}({e}_{m}/{e}_{2m}).

    Section 2 of this paper, confirms that the system is solvable for all q values in (0, 1) . In addition to that, both one-step [16] and R -step collocation estimate errors together with convergence rates, CPU time based on different delay parameters such as q = 0.2, 0.5, 0.9 , and various collocation points have been reported in the upcoming two test problems. We have demonstrated that both solvability and convergence of the system are achieved for all values of q in the interval (0, 1) . We have chosen a variety of q parameters. It has been compared with the one-step collocation method with the multistep collocation one. The results in both cases have been gathered and shown.

    As it can be seen through all tables, with nearly the same computational cost, the multistep collocation method achieves lower errors compared to the one-step collocation method.

    For the sake of comparing our results with the ones given in [1], we have provided a selection of the numerical example from [1].

    Example 1. [1] In the current test problem, one-step and multistep collocation-based techniques have been applied for the sake of solving the following pantograph DVIE:

    \begin{equation} x(T) = {g}_{1}(T)-\int_{0}^{T}x(\tau)\mathrm{d}\tau+\frac{1}{2}\int_{0}^{qT}x(\tau) \mathrm{d}\tau, \quad T\in J: = [0,10], \end{equation} (4.1)

    with considering the exact solution of x(T) = \exp(-T) , and

    \begin{equation*} {g}_{1}(T) = \frac{1}{2}\Big(1+\exp(-qT)\Big). \end{equation*}

    As noted, the convergence of one-step methods relies solely on the collocation parameters {c}_{i} . In contrast, multistep collocation methods depend not only on these parameters but also on the spectral radius of the coefficient matrix of the difference equation, denoted as \mathbf{S} .

    The proposed method relies on the parameters {c}_{i} , requiring that \rho(\mathbf{S}) < 1 for convergence to occur. It is important to note that article [1] reports the highest accuracy at the Gauss points; however, since the condition \rho(\mathbf{S}) < 1 is not satisfied at these points, our proposed numerical method does not converge there.

    For the sake of comparing with [1], we have employed the collocation parameters of {c}_{1} = 7/10 and {c}_{2} = 1 , with M = 2 , R = 3 , and q = 0.2, 0.5 . Then we have reached the accuracy of {10}^{-11} at N = 256 grid points, while at [1], it reaches the accuracy of {10}^{-11} for N = 1600 grid points.

    In the reference [1], the authors have exploited the quasi-geometric meshes, which are dependent on delay parameter q . As a result, for q = 0.9 , they have attained different solutions than the case, q = 0.2 and q = 0.5 . However, here the uniform meshes have been utilized, so we obtain the same results for all values of the parameter q .

    For the second test problem, we first take a monotone increasing function of x(T) = \sqrt{2+T} , which is totally different from an exponential function, as an exact solution, and then we have reported the results just for the delay parameter of q = 0.2 but for two selections of collocation parameters.

    Example 2. Herein, we have considered what follows.

    \begin{equation} x(T) = {g}_{2}(T)+\int_{0}^{T}{(T+\tau)}x(\tau)\mathrm{d}\tau+\int_{0}^{qT}Tx(\tau) \mathrm{d}\tau, \quad T\in J: = [0,1], \end{equation} (4.2)

    For which, we have carried out both classical and multistep collocation schemes by taking {g}_{2}(T) such that we have x(T) = \sqrt{2+T} .

    It is worthwhile to note that in the second test problem the integral interval is J: = [0, 1] , which differs from the one in the first example.

    All presented results in Tables 1-5 are strongly in agreement with the foregoing Theorem 3.1.

    Table 1.  Multistep and one-step collocation {L}_{\infty} -norm errors along with their corresponding rates for Eq (4.1) under uniform meshes, delay parameter q = 0.2 and collocation parameters \lbrace {c}_{1} = 7/10, {c}_{2} = 1\rbrace .
    Multistep CM One-step CM
    N {L}_{\infty}- error Rate CPU time(s) {L}_{\infty}- error Rate CPU time(s)
    32 3.7326E-07 - 2.250 2.2872E-03 - 2.953
    64 1.7579E-08 4.40 9.063 6.5226E-04 1.81 12.657
    128 6.7537E-10 4.70 38.969 1.6858E-04 1.95 56.874
    256 2.1412E-11 4.97 171.483 4.3716E-05 1.94 254.813

     | Show Table
    DownLoad: CSV
    Table 2.  Multistep and one-step collocation {L}_{\infty} -norm errors along with their corresponding rates for Eq (4.1) under uniform meshes, delay parameter q = 0.5 and collocation parameters \lbrace {c}_{1} = 7/10, {c}_{2} = 1\rbrace .
    Multistep CM One-step CM
    N {L}_{\infty}- error Rate CPU time(s) {L}_{\infty}- error Rate CPU time(s)
    32 3.7326E-07 - 2.718 2.1145E-03 - 2.609
    64 1.4659E-08 4.67 11.455 5.7019E-04 1.89 11.827
    128 5.2515E-10 4.80 48.937 1.4745E-04 1.95 48.906
    256 1.7739E-11 4.88 214.296 3.7528E-05 1.97 211.921

     | Show Table
    DownLoad: CSV
    Table 3.  Multistep and one-step collocation {L}_{\infty} -norm errors along with their corresponding rates for Eq (4.1) under uniform meshes, delay parameter q = 0.9 and collocation parameters \lbrace {c}_{1} = 7/10, {c}_{2} = 1\rbrace .
    Multistep CM One-step CM
    N {L}_{\infty}- error Rate CPU time(s) {L}_{\infty}- error Rate CPU time(s)
    32 2.4766E-07 - 4.296 1.8352E-03 - 4.485
    64 1.1967E-08 4.37 15.876 5.0747E-04 1.85 14.656
    128 4.6069E-10 4.69 64.922 1.3121E-04 1.95 62.720
    256 1.5832E-11 4.86 279.765 3.4008E-05 1.94 276.422

     | Show Table
    DownLoad: CSV
    Table 4.  Multistep and one-step collocation {L}_{\infty} -norm errors along with their corresponding rates for Eq (4.2) under uniform meshes, delay parameter q = 0.2 and collocation parameters \lbrace {c}_{1} = 7/10, {c}_{2} = 1\rbrace .
    Multistep CM One-step CM
    N {L}_{\infty}- error Rate CPU time(s) {L}_{\infty}- error Rate CPU time(s)
    8 2.0663E-09 - 0.437 3.2035E-04 - 0.374
    16 9.7556E-11 4.40 0.813 7.8609E-05 2.02 0.750
    32 4.0352E-12 4.59 2.157 1.9480E-05 2.01 2.156
    64 1.4210E-13 4.82 9.359 4.8319E-06 2.01 8.220

     | Show Table
    DownLoad: CSV
    Table 5.  Multistep and one-step collocation {L}_{\infty} -norm errors along with their corresponding rates for Eq (4.2) under uniform meshes, delay parameter q = 0.2 and collocation parameters \lbrace {c}_{1} = 1/2, {c}_{2} = 7/10, {c}_{3} = 1\rbrace .
    Multistep CM One-step CM
    N {L}_{\infty}- error Rate CPU time(s) {L}_{\infty}- error Rate CPU time(s)
    8 2.3015E-11 - 1.095 2.7042E-06 - 0.551
    16 6.8789E-13 5.06 1.860 3.3248E-07 3.02 0.907
    32 1.2434E-14 5.78 4.173 4.1104E-08 3.01 3.594
    64 2.1102E-16 5.88 12.76 5.0948E-09 3.01 10.23

     | Show Table
    DownLoad: CSV

    In this article, we employed a robust high-rate multistep collocation strategy in order to tackle second-kind vanishing delay VIEs. Both numerical and theoretical solution solvability have been proved detailedly. Moreover, a theorem associated with multistep collocation solution error analysis has been presented. In the last section, for the sake of confirming theoretically achieved consequences, two numerical examples were given, which have been investigated for a variety of delay parameters of q = 0.2, 0.5, 0.9 on uniform meshes perfectly. Furthermore, it is of great importance to remark that based on our study, high convergence rates for second-kind pantograph delay VIEs (1.1) can be attained for all various delay parameters of q in (0, 1) .

    Shireen Obaid Khaleel: Writing original draft, Visualization, Formal analysis; Parviz Darania: Writing original draft, Validation, Writing review & editing; Saeed Pishbin: Software, Validation, Writing review & editing; Shadi Malek Bagomghaleh: Writing original draft, Visualization, Formal analysis. All authors discussed the results and revised the draft. All authors have read and approved the final version of the manuscript for publication.

    The authors gratefully acknowledge the anonymous referees for their careful reading of the manuscript and constructive comments.

    All authors declare that they have no conflicts of interest.



    [1] D. C. Balmer, Impact of the A18. 1 ASME Standard on platform lifts and stairway chairlifts on accessibility and usability, Assist. Technol., 22 (2010), 46−50. https://doi.org/10.1080/10400430903520264 doi: 10.1080/10400430903520264
    [2] C. Cheng, S. Zhang, Z. Wang, L. Qiu, L. Tu, L. Zhu, et al., Surrogate-model-based dynamic-sensing optimization method for high-speed elevator car horizontal vibration reduction, Proc. I. Mech. Eng. Part C, 2024. https://doi.org/10.1177/09544062231217926 doi: 10.1177/09544062231217926
    [3] P. C. Liao, Z. Guo, T. Wang, J. Wen, C. H. Tsai, Interdependency of construction safety hazards from a network perspective: A mechanical installation case, Int. J. Occup. Saf. Ergo., 26 (2020), 245−255. https://doi.org/10.1080/10803548.2018.1426272 doi: 10.1080/10803548.2018.1426272
    [4] J. Lei, W. Sun, Y. Fang, N. Ye, S. Yang, J. Wu, A model for detecting abnormal elevator passenger behavior based on video classification, Electronics, 13 (2024), 2472. https://doi.org/10.3390/electronics13132472 doi: 10.3390/electronics13132472
    [5] H. Hasegawa, S. Aida, Elevator monitoring system to guide user's behavior by visualizing the state of crowdedness, In: Lee, R. (eds) Big Data, Cloud Computing, and Data Science Engineering, Springer, Cham, 2020, 85−98. https://doi.org/10.1007/978-3-030-24405-7_6
    [6] S. Liang, D. Niu, K. Huang, H. Wu, L. Ding, Y. Yang, An elevator door blocking behavior recognition method based on two-stage object detection networks, IEEE, 2022, 1374−1378. https://doi.org/10.1109/YAC57282.2022.10023898 doi: 10.1109/YAC57282.2022.10023898
    [7] Z. Wang, J. Chen, P. Yu, B. Feng, D. Feng, SC-YOLOv8 network with soft-pooling and attention for elevator passenger detection, Appl. Sci., 14 (2024), 3321. https://doi.org/10.3390/app14083321 doi: 10.3390/app14083321
    [8] S. Chai, X. I. Li, Y. Jia, Y. He, C. H. Yip, K. K. Cheung, et al., A non-intrusive deep learning based diagnosis system for elevators, IEEE Access, 9 (2021), 20993−21003. https://doi.org/10.1109/ACCESS.2021.3053858 doi: 10.1109/ACCESS.2021.3053858
    [9] S. C. Lai, M. L. Yang, R. J. Wang, J. Y. Jhuang, M. C. Ho, Y. C. Shiau, Remote-control system for elevator with sensor technology, Sensor. Mater., 34 (2022). https://doi.org/10.18494/SAM3827 doi: 10.18494/SAM3827
    [10] Z. Li, J. Ning, T. Li, Design of non-intrusive online monitoring system for traction elevators, Appl. Sci., 14 (2024), 4346. https://doi.org/10.3390/app14114346 doi: 10.3390/app14114346
    [11] W. Yao, A. Wang, Y. Nie, Z. Lv, S. Nie, C. Huang, et al., Study on the recognition of coal miners' unsafe behavior and status in the hoist cage based on machine vision, Sensors, 23 (2023), 8794. https://doi.org/10.3390/s23218794 doi: 10.3390/s23218794
    [12] T. Kong, W. Fang, P. E. D. Love, H. Luo, S. Xu, H. Li, Computer vision and long short-term memory: Learning to predict unsafe behaviour in construction, Adv. Eng. Inform., 50 (2021), 101400. https://doi.org/10.1016/j.aei.2021.101400 doi: 10.1016/j.aei.2021.101400
    [13] M. Casini, Extended reality for smart building operation and maintenance: A review, Energies, 15 (2022), 3785. https://doi.org/10.3390/en15103785 doi: 10.3390/en15103785
    [14] R. D'Souza, IoT and the future of elevator maintenance business, Master Thesis, Technische Universität Wien, 2022. https://doi.org/10.34726/hss.2022.103532
    [15] X. P. Zhang, J. H. Ji, L. Wang, Z. He, S. Liu, A review of video-based human abnormal behavior recognition and detection methods, Control Decis., 37 (2022), 14−27.
    [16] A. F. Bobick, J. W. Davis, The recognition of human movement using temporal templates, IEEE T. Pattern Anal., 23 (2001), 257−267. https://doi.org/10.1109/34.910878 doi: 10.1109/34.910878
    [17] H. Wang, A. Kläser, C. Schmid, C. L. Liu, Dense trajectories and motion boundary descriptors for action recognition, Int. J. Comput. Vis., 103 (2013), 60−79. https://doi.org/10.1007/s11263-012-0594-8 doi: 10.1007/s11263-012-0594-8
    [18] L. Xu, C. Gong, J. Yang, Q. Wu, L. Yao, Violent video detection based on MoSIFT feature and sparse coding, IEEE, 2014, 3538−3542. https://doi.org/10.1109/ICASSP.2014.6854259 doi: 10.1109/ICASSP.2014.6854259
    [19] H. Fujiyoshi, A. J. Lipton, T. Kanade, Real-time human motion analysis by image skeletonization, IEICE T. Inf. Syst., 87 (2004), 113−120.
    [20] M. S. Alzahrani, S. K. Jarraya, H. Ben-Abdallah, M. S. Ali, Comprehensive evaluation of skeleton features-based fall detection from Microsoft Kinect v2, Signal, Image Video P., 13 (2019), 1431−1439. https://doi.org/10.1007/s11760-019-01490-9 doi: 10.1007/s11760-019-01490-9
    [21] Z. Liao, H. Hu, J. Zhang, C. Yin, Residual attention unit for action recognition, Comput. Vis. Image Und., 189 (2019), 102821. https://doi.org/10.1016/j.cviu.2019.102821 doi: 10.1016/j.cviu.2019.102821
    [22] C. Feichtenhofer, A. Pinz, A. Zisserman, Convolutional two-stream network fusion for video action recognition, Proc. IEEE Conf. Comput. Vis. Pattern Recogn., 2016, 1933−1941. https://doi.org/10.1109/CVPR.2016.213 doi: 10.1109/CVPR.2016.213
    [23] S. Sudhakaran, O. Lanz, Learning to detect violent videos using convolutional long short-term memory, IEEE, 2017, 1−6. https://doi.org/10.1109/AVSS.2017.8078468 doi: 10.1109/AVSS.2017.8078468
    [24] C. Y. Wang, H. Y. M. Liao, Y. H. Wu, P. Y. Chen, J. W. Hsieh, I. H. Yeh, CSPNet: A new backbone that can enhance learning capability of CNN, Montreal, BC, Canada, 2020, 390−391.
    [25] X. Hu, D. Kong, X. Liu, J. Zhang, D. Zhang, FM-STDNet: High-speed detector for fast-moving small targets based on deep first-order network architecture, Electronics, 12 (2023), 1−15. https://doi.org/10.3390/electronics12081829 doi: 10.3390/electronics12081829
    [26] S. Wang, Z. Miao, Anomaly detection in crowd scene, IEEE, 2010, 1220−1223. https://doi.org/10.1109/ICOSP.2010.5655356 doi: 10.1109/ICOSP.2010.5655356
    [27] R. Mehran, A. Oyama, M. Shah, Abnormal crowd behavior detection using social force model, IEEE, 2009, 935−942. https://doi.org/10.1109/CVPR.2009.5206641 doi: 10.1109/CVPR.2009.5206641
    [28] J. Kim, K. Grauman, Observe locally, infer globally: A space-time MRF for detecting abnormal activities with incremental updates, IEEE, 2009, 2921−2928. https://doi.org/10.1109/CVPR.2009.5206569 doi: 10.1109/CVPR.2009.5206569
    [29] M. Hasan, J. Choi, J. Neumann, A. K. Roy-Chowdhury, L. S. Davis, Learning temporal regularity in video sequences, Proc. IEEE Conf. Comput. Vis. Pattern Recogn., 2016, 733−742. https://doi.org/10.1109/CVPR.2016.86 doi: 10.1109/CVPR.2016.86
    [30] W. Luo, W. Liu, S. Gao, Remembering history with convolutional lstm for anomaly detection, IEEE, 2017, 439−444. https://doi.org/10.1109/ICME.2017.8019325 doi: 10.1109/ICME.2017.8019325
    [31] Y. S. Chong, Y. H. Tay, Abnormal event detection in videos using spatiotemporal autoencoder, Springer, Cham, 2017, 189−196. https://doi.org/10.1007/978-3-319-59081-3_23
    [32] Q. Sun, H. Liu, T. Harada, Online growing neural gas for anomaly detection in changing surveillance scenes, Pattern Recogn., 64 (2017), 187−201. https://doi.org/10.1016/j.patcog.2016.09.016 doi: 10.1016/j.patcog.2016.09.016
    [33] M. Ravanbakhsh, M. Nabi, E. Sangineto, L. Marcenaro, C. Regazzoni, N. Sebe, Abnormal event detection in videos using generative adversarial nets, IEEE, 2017, 1577−1581. https://doi.org/10.1109/ICIP.2017.8296547 doi: 10.1109/ICIP.2017.8296547
    [34] R. Hinami, T. Mei, S. Satoh, Joint detection and recounting of abnormal events by learning deep generic knowledge, Proc. IEEE Conf. Comput. Vis., 2017, 3619−3627. https://doi.org/10.1109/ICCV.2017.391 doi: 10.1109/ICCV.2017.391
    [35] R. I. Tudor, S. Smeureanu, B. Alexe, M. Popescu, Unmasking the abnormal events in video, Proc. IEEE Conf. Comput. Vis., 2017, 2895−2903.
    [36] W. Luo, W. Liu, S. Gao, A revisit of sparse coding based anomaly detection in stacked RNN framework, Proc. IEEE Conf. Comput. Vis., 2017, 341−349. https://doi.org/10.1109/ICCV.2017.45 doi: 10.1109/ICCV.2017.45
    [37] W. Liu, W. Luo, D. Lian, S. Gao, Future frame prediction for anomaly detection—a new baseline, Proc. IEEE Conf. Comput. Vis. Pattern Recogn., 2018, 6536−6545. https://doi.org/10.1109/CVPR.2018.00684 doi: 10.1109/CVPR.2018.00684
    [38] D. Xu, Y. Yan, E. Ricci, N. Sebe, Detecting anomalous events in videos by learning deep representations of appearance and motion, Comput. Vis. Image Und., 156 (2017), 117−127. https://doi.org/10.1016/j.cviu.2016.10.010 doi: 10.1016/j.cviu.2016.10.010
    [39] M. Ravanbakhsh, M. Nabi, H. Mousavi, E. Sangineto, N. Sebe, Plug-and-play cnn for crowd motion analysis: An application in abnormal event detection, IEEE, 2018, 1689−1698. https://doi.org/10.1109/WACV.2018.00188 doi: 10.1109/WACV.2018.00188
    [40] R. T. Ionescu, F. S. Khan, M. I. Georgescu, L. Shao, Object-centric auto-encoders and dummy anomalies for abnormal event detection in video, 2019, 7842−7851. https://doi.org/10.1109/CVPR.2019.00803
    [41] C. Sun, Y. Jia, H. Song, Y. Wu, Adversarial 3d convolutional auto-encoder for abnormal event detection in videos, IEEE T. Multimedia, 23 (2020), 3292−3305. https://doi.org/10.1109/TMM.2020.3023303 doi: 10.1109/TMM.2020.3023303
  • This article has been cited by:

    1. Mostafa M.A. Khater, Mustafa Inc, K.S. Nisar, Raghda A.M. Attia, Multi–solitons, lumps, and breath solutions of the water wave propagation with surface tension via four recent computational schemes, 2021, 20904479, 10.1016/j.asej.2020.10.029
    2. Mostafa M. A. Khater, Yu-Ming Chu, Raghda A. M. Attia, Mustafa Inc, Dianchen Lu, On the Analytical and Numerical Solutions in the Quantum Magnetoplasmas: The Atangana Conformable Derivative (1+3)-ZK Equation with Power-Law Nonlinearity, 2020, 2020, 1687-9120, 1, 10.1155/2020/5809289
    3. Usman Younas, Muhammad Bilal, Tukur Abdulkadir Sulaiman, Jingli Ren, Abdullahi Yusuf, On the exact soliton solutions and different wave structures to the double dispersive equation, 2022, 54, 0306-8919, 10.1007/s11082-021-03445-2
    4. Ghazala Akram, Maasoomah Sadaf, Iqra Zainab, Observations of fractional effects of β-derivative and M-truncated derivative for space time fractional Phi-4 equation via two analytical techniques, 2022, 154, 09600779, 111645, 10.1016/j.chaos.2021.111645
    5. Mojtaba Shahraki, Hossein Malekpoor, The effects of polarization on soliton interactions inside optical fibers, 2022, 54, 0306-8919, 10.1007/s11082-022-03514-0
    6. Md. Abdul Kayum, Ripan Roy, M. Ali Akbar, M. S. Osman, Study of W-shaped, V-shaped, and other type of surfaces of the ZK-BBM and GZD-BBM equations, 2021, 53, 0306-8919, 10.1007/s11082-021-03031-6
    7. Khalid K. Ali, Hadi Rezazadeh, Nauman Raza, Mustafa Inc, Highly dispersive optical soliton perturbation with cubic–quintic–septic law via two methods, 2021, 35, 0217-9792, 10.1142/S0217979221502763
    8. Delmar Sherriffe, Diptiranjan Behera, P. Nagarani, Different forms for exact traveling wave solutions of unstable and hyperbolic nonlinear Schrödinger equations, 2024, 38, 0217-9792, 10.1142/S0217979224501315
    9. Handenur Esen, Neslihan Ozdemir, Aydin Secer, Mustafa Bayram, Tukur Abdulkadir Sulaiman, Hijaz Ahmad, Abdullahi Yusuf, M. Daher Albalwi, On the soliton solutions to the density-dependent space time fractional reaction–diffusion equation with conformable and M-truncated derivatives, 2023, 55, 0306-8919, 10.1007/s11082-023-05109-9
    10. Hira Tariq, Hira Ashraf, Hadi Rezazadeh, Ulviye Demirbilek, Travelling wave solutions of nonlinear conformable Bogoyavlenskii equations via two powerful analytical approaches, 2024, 39, 1005-1031, 502, 10.1007/s11766-024-5030-7
    11. Rui Cui, New waveform solutions of Calogero–Degasperis (CD) and potential Kadomtsev–Petviashvili (pKP) equations, 2024, 7, 2520-8160, 1673, 10.1007/s41939-023-00254-w
  • Reader Comments
  • © 2024 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(628) PDF downloads(36) Cited by(0)

Figures and Tables

Figures(12)  /  Tables(3)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog