Multi-target Tracking Problem Research Articles

CLONE: Collaborative Learning on the Edges

The proliferation of edge computing technologies has boosted the development of new applications for a plethora of edge devices. However, many applications face privacy issues and bandwidth limitations. To solve these limitations, we propose a collaborative learning framework on the edges, named CLONE, which is steered by the real-world data sets collected from a large electric vehicle (EV) company and a grocery store of a shopping mall, respectively. We categorize two application scenarios for CLONE, i.e., CLONE in the training stage (CLONE_training) and CLONE in the inference stage (CLONE_inference). As to CLONE_training, we choose the failure prediction of EV battery and associated components as the first use case. While as for CLONE_inference, customer tracking in a grocery store is selected as another case study. In this work, the goal of the CLONE is to support real-time training and inference for connected vehicles and marketing intelligence services. Our experimental results on the EV data show that CLONE is able to reduce model training time without sacrificing algorithm performance. Furthermore, the experimental results on the video data from the grocery store reveal that CLONE is a useful approach to solve the multitarget multicamera tracking problem in a collaborative fashion.

Finite-time multi-target tracking of networked Euler–Lagrange systems via hierarchical active disturbance rejection control

This paper investigates the finite-time multi-target tracking problem of networked Euler–Lagrange systems (NELSs) with a directed interaction topology, parametric uncertainties and external disturbances. A novel finite-time hierarchical active disturbance rejection control (ADRC) algorithm is designed to handle such a challenging problem without velocity measurement. To be specific, a nonsingular distributed estimator algorithm is newly designed to respectively estimate the states of the targets in a finite time. Based on the extended state observer (ESO), we present a finite-time local control scheme to obtain satisfactory tracking performance of the NELSs. Besides, sufficient conditions for deriving the finite-time stability of the cascade closed-loop systems are presented based on the Lyapunov argument. Finally, numerical simulations are performed to show the control performance and feasibility of the main results.

Multitarget tracking control algorithm under local information selection interaction mechanism

This study focuses on the problem of multitarget tracking. To address the existing problems of current tracking algorithms, as manifested by the time consumption of subgroup separation and the uneven group size of unmanned aerial vehicles (UAVs) for target tracking, a multitarget tracking control algorithm under local information selection interaction is proposed. First, on the basis of location, number, and perceived target information of neighboring UAVs, a temporary leader selection strategy is designed to realize the local follow-up movement of UAVs when the UAVs cannot fully perceive the target. Second, in combination with the basic rules of cluster movement and target information perception factors, distributed control equations are designed to achieve a rapid gathering of UAVs and consistent tracking of multiple targets. Lastly, the simulation experiments are conducted in two- and three-dimensional spaces. Under a certain number of UAVs, clustering speed of the proposed algorithm is less than 3s, and the equal probability of the UAV subgroup size after group separation is over 78%.

Adaptive probability hypothesis density filter for multi-target tracking with unknown measurement noise statistics

Under the Gaussian noise assumption, the probability hypothesis density (PHD) filter represents a promising tool for tracking a group of moving targets with a time-varying number. However, inaccurate prior statistics of the random noise will degrade the performance of the PHD filter in many practical applications. This paper presents an adaptive Gaussian mixture PHD (AGM-PHD) filter for the multi-target tracking (MTT) problem in the scenario where both the mean and covariance of measurement noise sequences are unknown. The conventional PHD filters are extended to jointly estimate both the multi-target state and the aforementioned measurement noise statistics. In particular, the Normal-inverse-Wishart and Gaussian distributions are first integrated to represent the joint posterior intensity by transforming the measurement model into a new formulation. Then, the updating rule for the hyperparameters of the model is derived in closed form based on variational Bayesian (VB) approximation and Bayesian conjugate prior heuristics. Finally, the dynamic system state and the noise statistics are updated sequentially in an iterative manner. Simulations results with both constant velocity and constant turn model demonstrate that the AGM-PHD filter achieves comparable performance as the ideal PHD filter with true measurement noise statistics.

A Multi Target Algorithm for Image Target Tracking

As technology advances, improving the accuracy of image target tracking is extremely important for medicine, aerospace and military, industry and agriculture. Image target tracking usually apply the method of manual extraction, which is not only a waste of time and power, but also has many difficulties in extraction. However, multi target algorithm can make up for its shortcomings, and multi target algorithm develops from underlying characteristics to multi-level, abstract features in the form of data. This paper, proposed a novel multi target algorithm for image target tracking for solving the above shortcomings. In the multi-target tracking problem, the random set is actually the number of elements and a set of random variables. When the number of targets is unknown or changing, the number of targets is a discrete random variable, the dimension of the state space will be different with the target value of the number of changes, so the multi target state model and observation model can be expressed as random finite sets form. Simulation results show that compared with traditional method, this method has higher accuracy and low loss rate.

A Bernoulli Track-Before-Detect Filter for Interacting Targets in Maritime Radar

This article is devoted to the problem of multitarget tracking in the maritime environment using Bernoulli track-before-detect (TBD) filtering. Detection and tracking of weak targets in sea clutter using high-resolution airborne radar is notoriously challenging compared to classical target tracking problems. Furthermore, the measured amplitude in range-azimuth cells that are in the vicinity of closely spaced (interacting) targets represents a superposition of individual target contributions, demanding a special joint treatment. This article proposes a novel solution for multitarget tracking using Bernoulli TBD for maritime radar and covers cases when targets are both far apart and closely spaced. Comparisons with existing TBD algorithms are used to verify the efficacy of the proposed Bernoulli TBD algorithm with realistic sea clutter simulations used to create challenging scenarios with variable target strengths.

A Multilane Tracking Algorithm Using IPDA with Intensity Feature.

Detection of multiple lane markings on road surfaces is an important aspect of autonomous vehicles. Although a number of approaches have been proposed to detect lanes, detecting multiple lane markings, particularly across a large number of frames and under varying lighting conditions, in a consistent manner is still a challenging problem. In this paper, we propose a novel approach for detecting multiple lanes across a large number of frames and under various lighting conditions. Instead of resorting to the conventional approach of processing each frame to detect lanes, we treat the overall problem as a multitarget tracking problem across space and time using the integrated probabilistic data association filter (IPDAF) as our basis filter. We use the intensity of the pixels as an augmented feature to correctly group multiple lane markings using the Hough transform. By representing these extracted lane markings as splines, we then identify a set of control points, which becomes a set of targets to be tracked over a period of time, and thus across a large number of frames. We evaluate our approach on two different fronts, covering both model- and machine-learning-based approaches, using two different datasets, namely the Caltech and TuSimple lane detection datasets, respectively. When tested against model-based approach, the proposed approach can offer as much as , , and improvements on the true positive, false positive, and false positives per frame rates compared to the best alternative approach, respectively. When compared against a state-of-the-art machine learning technique, particularly against a supervised learning method, the proposed approach offers , , , and improvements on the false positive, false negative, accuracy, and frame rates. Furthemore, the proposed approach retains the explainability, or in other words, the cause of actions of the proposed approach can easily be understood or explained.

Adaptive Non-Linear Joint Probabilistic Data Association for Vehicle Target Tracking

This paper investigated the problem of multi-target tracking (MTT) over a vehicle sensor network. A novel adaptive square root cubature joint probabilistic data association (ASRCJPDA) was proposed. Motivated by enhancing the stability of joint probabilistic data association (JPDA) in practical application, the proposed methodology implemented a numerically stabled cubature Kalman filter for JPDA state estimate process. It improved numerical stability and acquired more accurate estimated results. Additionally, enlightened by enhancing the real time efficiency of JPDA, an adaptive tracking gate designed for the JPDA measurement associate process was proposed. It combined with the kinematics of vehicle to reduce the computational complexity of data association, which improved the robustness of MTT in complex scenarios. The virtual vehicle target tracking scenarios were built in PreScan software in order to better simulate the real traffic condition. Simulations of target tracking examples are presented to show great effectiveness and superiority of the proposed method.

Spatial-temporal constrained particle filter for cooperative target tracking

Localization is one of the most important topics of cyber physical system. In the last decades, much attention has been paid to the precise localization and performance evaluation in wireless sensor networks. Inertial-measurement-unit and Time-of-Arrival fusion is a state-of-the-art method to solve the accumulative error and drifting problem faced by sole inertial-measurement-unit positioning and navigation. Network cooperative technology could effectively suppress the accumulative error. This paper presents a spatial-temporal constrained particle filter algorithm for cooperative target tracking, so as to solve the problem of multi-target high-precision position tracking in complex and highly dynamic environments. Firstly, we propose an error-ellipse-resampling particle filter method. In the resampling stage of the particle filter, error ellipses with different confidence probabilities would be established with the use of the known estimated center and confidence scale, to achieve hierarchical resampling optimization based on the geometrical position of particles. As for cooperative tracking, an optimization method of state estimation based on spatial distance constraint is proposed, so that the Bayesian filter can benefit from spatial information and achieve cooperative tracking of spatial-temporal information fusion. Numerical experimental results show that the proposed error-ellipse-resampling particle filter could decrease the growth rate of cumulative errors and reach a positioning accuracy of 1.05 m, multi-target cooperative error-ellipse-resampling particle filter can effectively eliminate the cumulative error and achieve a positioning accuracy of 0.24 m.

Multi-Sensor Fusion Tracking Algorithm Based on Augmented Reality System

With the rapid development of modern science and technology, multi-sensor information fusion target tracking technology has gradually become an increasingly important topic in the field of computer vision. It is not only of great significance in traffic monitoring, unmanned vehicle systems, human-computer interaction, etc., but also has a wide range of applications in target tracking. Aiming at the problem of multi-sensor multi-target tracking, this article proposes a multi-sensor multi-target tracking algorithm based on random finite set. The basic theory of random sets, random finite set statistics and the probability hypothesis density of random sets are introduced. The distribution function of random sets and the derivation process of set derivatives and set integrals are discussed, and the theoretical framework of information fusion based on random sets is given. The single-sensor single-target specification Bayes modeling method is extended to the multi-sensor multi-target situation, and the significance of the probability hypothesis density is explained, and the multi-sensor multi-target tracking is realized by the probability hypothesis density filtering. The simulation experiment results show that the algorithm has a better tracking effect for the maneuvering target in the nonlinear motion model stage, and it is verified that the unscented Kalman filter has smaller filtering error and higher estimation accuracy than the extended Kalman filter.

A robust fast LMB filter for superpositional sensors

This paper aims to deal with the problem of multitarget tracking based on superpositional sensors, while the measurement noise is unknown. To this end, the labeled multi-Bernoulli (LMB) filter is adopted in order to jointly estimate the number of existing targets and the state of each target recursively. Since the LMB filter admits no analytical solution, a Gaussian mixture (GM) approximated implementation is proposed. In order to deal with the unknown measurement noise, the single target state is augmented to incorporate the noise parameters, and the augmented state is modeled as the Gaussian and inverse Wishart mixture (GIWM) distribution. The posterior is approximately computed using the variational Bayesian (VB) estimation method so as to ensure that the updated distributions of augmented states belong to the same family of the predicted one, thus can be used as prior for the subsequent iterations. The performance of the proposed algorithm is assessed via simulation experiments.

An efficient implementation of multiple weak targets tracking filter with labeled random finite sets for marine radar

The detection and tracking of multiple weak targets with time-varying and unknown number has become a hot spot and challenging problem in marine radar application. This paper integrates δ-generalized labeled multi-Bernoulli (δ-GLMB) density and labeled multi-Bernoulli (LMB) density, which are two important special cases of labeled random finite set, to propose an effective method for multiple weak target tracking with track-before-detect strategy based on Bayes framework including prediction and update step. The proposed method can deal with multi-target tracking problem with tractable computational complexity due to prediction step with dynamic grouping procedure albeit with slightly precision loss compared with standard δ-GLMB filter. In general, the conjugate prior property based GLMB does not hold on for generic observation model in update step. In order to solve this problem, this paper applies a tractable principled approximation involving GLMB density, which was firstly proposed in the literature, to make whole Bayes filtering with labeled RFS to perform iteratively for each group with TBD model. Finally, the numerical simulation results of Swerling type 1 target tracking in K-distributed sea clutter illustrate that the proposed method is superior to the methods including standard δ-GLMB-TBD filter, LMB-TBD filter and DP-TBD algorithm in comprehensive performance.

Gaussian mixture probability hypothesis density filter against measurement origin uncertainty

The Gaussian mixture probability hypothesis density (GM-PHD) filter is a promising solution to the multi-target tracking (MTT) problem, which successfully integrates target detection, tracking, and identification. Despite its wide applicability and computational efficiency, the existing GM-PHD filter can lose the estimates of the targets frequently in heavily cluttered and/or low signal-to-noise ratio (SNR) environments. This is mainly attributed to insufficient consideration of uncertainties around whether a measurement is from a target or not in the GM-PHD filter. Specifically, at each time step, the GM-PHD filter generates new Gaussian components corresponding to individual measurements which have the same estimate error covariances regardless of whether the measurement is from a target or not, so that it can lose the estimates of targets when the clutter density is high and/or the detection probability is low. To address this problem, a new covariance update equation is proposed. This equation computes the estimate error covariance of a newly generated Gaussian component corresponding to each measurement conditioned on the uncertainty in the measurement origin, i.e. whether: (1) measurement is clutter; (2) measurement is originated from a target; and (3) there is no measurement. The performance of the proposed GM-PHD algorithm is demonstrated with illustrative MTT scenarios with different noise conditions.

Decentralized Gaussian Filters for Cooperative Self-Localization and Multi-Target Tracking

Scalable and decentralized algorithms for Cooperative Self-localization (CS) of agents, and Multi-Target Tracking (MTT) are important in many applications. In this work, we address the problem of Simultaneous Cooperative Self-localization and Multi-Target Tracking (SCS-MTT) under target data association uncertainty, i.e., the associations between measurements and target tracks are unknown. Existing CS and tracking algorithms either make the assumption of no data association uncertainty or employ a hard-decision rule for measurement-to-target associations. We propose a novel decentralized SCS-MTT method for an unknown and time-varying number of targets under association uncertainty. Marginal posterior densities for agents and targets are obtained by an efficient belief propagation (BP) based scheme while data association is handled by marginalizing over all target-to-measurement association probabilities. Decentralized single Gaussian and Gaussian mixture implementations are provided based on average consensus schemes, which require communication only with one-hop neighbors. An additional novelty is a decentralized Gibbs mechanism for efficient evaluation of the product of Gaussian mixtures. Numerical experiments show the improved CS and MTT performance compared to the conventional approach of separate localization and target tracking.

Joint Target Detection and Tracking in Multipath Environment: A Variational Bayesian Approach

We consider multitarget detection and tracking problem for a class of multipath detection system where one target may generate multiple measurements via multiple propagation paths, and the association relationship among targets, measurements, and propagation paths is unknown. In order to effectively utilize multipath measurements from one target to improve detection and tracking performance, a tracker has to handle high-dimensional estimation of latent variables including target existence state, target kinematic state, and multipath data association. Based on variational Bayesian inference, we propose a novel joint detection and tracking algorithm that incorporates multipath data association, target detection, and target state estimation in a unified Bayesian framework. The posterior probabilities of these latent variables are derived in a closed-form iterative manner, which is effective for reducing the performance deterioration caused by the coupling between estimation errors and identification errors. Loopy belief propagation is exploited to approximately calculate the probability of multipath data association, saving the computational cost significantly. Simulation results of over-the-horizon radar multitarget tracking show that the proposed algorithm outperforms multihypothesis multipath track fusion and multidetection (hypothesis-oriented) multiple hypothesis tracker, especially under low signal-to-noise ratio circumstance.

Adaptive cardinality balanced multi‐target multi‐Bernoulli filter based on cubature Kalman

The sequential Monte Carlo cardinality balanced multi-Bernoulli (SMC-CBMeMBer) filter provides a good framework to cope with the multi-target tracking problem. However, the standard SMC-CBMeMBer filter suffers from the particles’ degradation problem seriously. Using the measurements to construct the proposal density in the step of predict can effectively solve the above problem, but this kind of approach brings an amount of computation and causes the overestimation of the target number. To examine the quality of each predicted particle adaptively and use the cubature Kalman filter (CKF) to refine the poor-quality particles with the aid of the current measurements is proposed in this study. This method manages to alleviate the particles degradation problem without increasing the computational complexity seriously since only a part of the particles is refined by the CKF. Also, the proposed method can avoid cardinality overestimation caused by abuse of measurements. A range of simulations is performed to test the performance of the proposed method. The results confirm the effectiveness and robustness of the novel method.

Multi-target tracking of networked heterogeneous collaborative robots in task space

This paper investigates the multi-target tracking problem of networked heterogeneous collaborative robots with parametric uncertainties and external disturbances in the task space, where each robot can be kinematic redundant or non-redundant. The acceleration of the targets can either belong to $${\mathcal {L}}_2$$ or $${\mathcal {L}}_{\infty }$$ space. A uniform distributed controller–estimator algorithm is designed to solve the aforementioned problem. The sufficient conditions for both zero-error stability and Lagrange stability of the closed-loop system are derived. Finally, simulation examples are illustrated to demonstrate the effectiveness of the main results.

Efficient box particle implementation of the multi-sensor GLMB filter in the presence of triple measurement uncertainty

The multi-sensor generalized labeled multi-Bernoulli (MSGLMB) filter is a tractable solution to multi-sensor multi-target tracking problems. The particle MSGLMB filter, which is time-consuming, can be used for the situation where some sensors return point measurements, while the others return box measurements. The box particle MSGLMB filter, which is potentially time-saving, can also be used for the situation if the point measurements are first transformed into boxes by their three-sigma supports. However, such box particle MSGLMB filter can be inefficient due to the rapidly multiplying contracted versions of box particles during multi-sensor update especially when the number of sensors or measurements therefrom is large. To reduce the inefficiency, we seek to reduce the number of contracted versions by processing point measurements directly with a new likelihood, namely the likelihood of a point measurement given a box particle, which does not require point measurements to contract box particles. Although an exact form of the said likelihood is unobtainable, two crude approximations of it are derived and studied, based on which an efficient box particle implementation for the situation is contributed. Simulation results prove that the proposed approach is effective.

A grey wolf optimizer using Gaussian estimation of distribution and its application in the multi-UAV multi-target urban tracking problem

To overcome premature convergence in the grey wolf optimizer (GWO), in this study, a modified GWO integrating the basic GWO with the Gaussian estimation of distribution (GED) strategy, called GEDGWO, is proposed. GEDGWO employs the Gauss probability model to estimate the distribution of the selected superior individuals and shifts the weighted mean to adjust the search directions. Additionally, a Gaussian distribution based inferior solutions repair (ISR) method is introduced to modify the ill-shaped distribution of the population. A disturbed Gaussian random walk method is utilized to strengthen the local exploration ability. The performance of GEDGWO is compared with those of other promising GWO variants and state-of-the-art algorithms on a benchmarking CEC 2014 test suite. Non-parametric Wilcoxon and Friedman tests as well as the post hoc Iman–Davenport test are performed to further verify the efficacy of GEDGWO. Moreover, GEDGWO is applied to solve multi-UAV multi-target urban tracking path planning problem. To overcome the shortcomings of the previous solution model, a new model is described to address this complex real-time engineering optimization problem. The validity and practicability of the problem models as well as the accuracy and efficiency of GEDGWO are demonstrated by the experimental results.

Robust multi-target tracking under mismatches in both dynamic and measurement models

In multi-target tracking (MTT), it is hard to guarantee that the dynamic and measurement model mismatches are described by process and measurement noises with suitable covariance matrices (CMs). Meanwhile, model changes caused by target maneuvers and disturbance-corrupted measurements may coexist in a complex environment. Therefore, we propose the problem of MTT under mismatches in both dynamic and measurement models, where the uncertain state prediction error and measurement noise CMs are considered as random variables. Then, a robust generalized labeled multi-Bernoulli (RGLMB) filter framework is derived to recursively propagate the joint posterior density of the multi-target state and the random CMs. By modeling the random CMs as inverse Wishart distributions, on-line random CMs tuning and multi-target state estimation are unified via minimizing the Kullback–Leibler divergence. Simulation results show that the proposed RGLMB filter is robust to different target motion models and noisy radar measurements corrupted by step, sinusoidal and stochastic forms of bias.