Multiple Target Tracking Research Articles

Bounded rational Dubins vehicle coordination for target tracking using reinforcement learning

In this paper, we address the problem of cooperative tracking of multiple heterogeneous targets by deploying multiple and heterogeneous pursuers exhibiting different decision-making capabilities. Initially, under infinite resources, we formulate a game between the evader and the pursuing team, with an evader being the maximizing player and the pursuing team being the minimizing one. Subsequently, we relax the perfect rationality assumption via the use of a level-k thinking framework that allows the evaders to not exhibit the same levels of rationality. Such rationality policies are computed by using a reinforcement learning-based architecture and are proven to form Nash policies as the thinking levels increase. Finally, in the case of multiple pursuers against multiple targets, we develop a switched learning scheme with multiple convergence sets by assigning the most intelligent pursuers to the most intelligent evaders.

Joint Strategy of Power and Bandwidth Allocation for Multiple Maneuvering Target Tracking in Cognitive MIMO Radar With Collocated Antennas

In this article, we propose a joint strategy of power and bandwidth allocation (JSPBA) for multiple maneuvering target tracking (MMTT) in the multi-input multi-output (MIMO) radar with collocated antennas. The basis of our strategy is to optimally allocate the transmitted resources of power and effective bandwidth by the prior information in the closed-loop system of cognitive tracking. On account of the predicted conditional Cramér-Rao lower bound (PC-CRLB) offering a more accurate and time-sensitive lower bound than the standard posterior CRLB (PCRLB), the PC-CRLB of the range, Doppler frequency, and direction-of-arrival (DOA) is derived, normalized and adopted as the optimization criterion. Moreover, in order to solve the nonconvex problem, the initial nonconvex problem is converted into a series of convex problems, which are further formulated as the standard semi-definite programming (SDP) problems and then be solved, by introducing the convex relaxation technique and the two-step solution technique. Simulations confirm the superiority of the proposed JSPBA algorithm, in terms of the overall tracking accuracy in the MMTT scenario.

Labeled Multi-Bernoulli Filter Based Multiple Resolvable Group Targets Tracking With Leader-Follower Model

This paper addresses the multiple resolvable group targets tracking (MRGTT) problem wherein the leader-follower (LF) model is adopted to evolve the group structure. Since there is an interaction between the destination and each leader, such that an attraction force of the destination information on the leader needs to be considered. To cope with the time-varying number of targets/groups and target-measurement association uncertainty, the so-called labeled multi-Bernoulli (LMB) filter is regarded as the filtering tool and combined with the LF model, named LF-LMB. Totally speaking, the implementation of the proposed LF-LMB filter consists of three steps. Firstly, the stochastic differential equation (SDE) of the state is used to, respectively, characterize the motion models of the leader and the follower. Then, the state transition equations (STEs) for the leader and the follower are derived. Lastly, the derived STEs are used to modify the prediction process of the standard LMB filter. Simulation experiments are carried out to demonstrate the superiority (e.g., optimal sub-pattern assignment (OSPA) and OSPA <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^{(2)}$</tex-math></inline-formula> errors are reduced by at least 15 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> comparing to the standard LMB filter) and robustness of the proposed approach.

A Biologically Inspired Appearance Modeling and Sample Feature-based Approach for Visual Target Tracking in Aerial Images

Visual tracking in uncrewed aerial vehicles is challenging because of the target appearance. Various research has been fulfilled to overcome appearance variations and unpredictable moving target issues. Visual saliency-based approaches have been widely studied in biologically inspired algorithms to detect moving targets based on attentional regions (ARs) extraction. This paper proposes a novel visual tracking method to deal with these issues. It consists of two main phases: spatiotemporal saliency-based appearance modeling (SSAM) and sample feature-based target detection (SFTD). The proposed method is based on a tracking-by-detection approach to provide a robust visual tracking system under appearance variation and unpredictable moving target conditions. Correspondingly, a semi-automatic trigger-based algorithm is proposed to handle the phases' operation, and a discriminative-based method is utilized for appearance modeling. In the SSAM phase, temporal saliency extracts the ARs and coarse segmentation. Spatial saliency is utilized for the object’s appearance modeling and spatial saliency detection. Because the spatial saliency detection process is time-consuming for multiple target tracking conditions, an automatic algorithm is proposed to detect the region saliences in a multithreading implementation that leads to low processing time. Consequently, the temporal and spatial saliencies are integrated to generate the final saliency and sample features. The generated sample features are transferred to the sample feature-based target detection (SFTD) phase to detect the target in different images based on samples. Experimental results demonstrate that the proposed method is effective and presents promising results compared to other existing methods.

Motion Coordination of Multi-Agent Networks for Multiple Target Tracking with Guaranteed Collision Avoidance

Motion Coordination of Multi-Agent Networks for Multiple Target Tracking with Guaranteed Collision Avoidance

A Microvision-Based Motion Measurement System for Nanopositioners Using the Feature-to-Phase Method

The development of automation at the nanoscale has been calling for precision motion sensing for robotic nanopositioners. This study presents a microvision-based measurement system for the accurate and efficient motion sensing of three-degree-of-freedom (DOF) nanopositioners. In this measurement system, an optimized target searching (OTS) scheme is proposed for automatic tracking target selection. A strategy is designed to combine feature matching with phase correlation to balance the measurement accuracy and efficiency. By proposing a multiple target tracking scheme, high-precision angular measurement is achieved, and velocity estimation is also provided in this microvision-based measurement system. Subsequently, offline simulations and online experiments are performed to comprehensively evaluate the performances of the microvision-based system with comparisons to traditional instrumentation. The simulation and experimental results demonstrate that the proposed system and marker-free method can realize excellent extensibility, strong robustness, and high precision of motion tracking for the nanopositioners, with measurement accuracy higher than 93 nm and 96 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mu $ </tex-math></inline-formula> rad.

Siam-Sort: Multi-Target Tracking in Video SAR Based on Tracking by Detection and Siamese Network

Shadows are widely used in the tracking of moving targets by video synthetic aperture radar (video SAR). However, they always appear in groups in video SAR images. In such cases, track effects produced by existing single-target tracking methods are no longer satisfactory. To this end, an effective way to obtain the capability of multiple target tracking (MTT) is in urgent demand. Note that tracking by detection (TBD) for MTT in optical images has achieved great success. However, TBD cannot be utilized in video SAR MTT directly. The reasons for this is that shadows of moving target are quite different from in video SAR image than optical images as they are time-varying and their pixel sizes are small. The aforementioned characteristics make shadows in video SAR images hard to detect in the process of TBD and lead to numerous matching errors in the data association process, which greatly affects the final tracking performance. Aiming at the above two problems, in this paper, we propose a multiple target tracking method based on TBD and the Siamese network. Specifically, to improve the detection accuracy, the multi-scale Faster-RCNN is first proposed to detect the shadows of moving targets. Meanwhile, dimension clusters are used to accelerate the convergence speed of the model in the training process as well as to obtain better network weights. Then, SiamNet is proposed for data association to reduce matching errors. Finally, we apply a Kalman filter to update the tracking results. The experimental results on two real video SAR datasets demonstrate that the proposed method outperforms other state-of-art methods, and the ablation experiment verifies the effectiveness of multi-scale Faster-RCNN and SimaNet.

Gaussian mixture probability hypothesis density filter with dynamic probabilities: Application to road traffic surveillance

With the development of autonomous driving system, accurate tracking of the number of vehicles and their states is critical to ensure traffic safety. Since the numbers and states of the vehicles are unknown and time-varying, the road traffic surveillance can be considered as a multiple-target tracking (MTT) problem. The Gaussian mixture probability hypothesis density (GM-PHD) filter is a popular approach in solving MTT problem and has been successfully applied to a range of applications including ground vehicle tracking. However, due to the road constraint, the maneuverability of a ground vehicle is limited to the road lane. Thus, the target detection probability, survival probability, and birth rate, which are assumed to be constant in the original GM-PHD filter, vary according to the surveillance area as well as its own state. To address this, we propose an algorithm that dynamically updates the probability of detection and survival, and birth rate simultaneously in a mathematically rigorous way. This is done by integrating the semantic information inferred from the environment with the state-dependent GM-PHD filter. The performance and effectiveness of the proposed algorithm are demonstrated via illustrative road traffic scenarios.

Fully Adaptive Radar for Multiple Target Tracking

This work focuses on the development of a fully adaptive radar multiple target tracking (FAR-MTT) model and demonstrates its benefit over static parameter selections. The FAR-MTT framework is an extension of the fully adaptive radar (FAR) framework for single-target tracking (STT). In transitioning FAR-STT concepts to FAR-MTT, the focus was on developing a multiple target Fisher information matrix to capture the effects of multiple targets on the radar measurements, and on developing a robust but straightforward optimization scheme/objective function in the FAR executive processor. Simulation and experimental examples are presented to demonstrate the performance of the FAR-MTT approach. Compared to the fixed parameter case, the FAR-MTT technique is capable of obtaining specified tracking performance for each target and ensuring target separation while reducing sensor resource usage in a variety of multiple target environments. In Monte Carlo testing the method resulted in track merges 28% of the time compared with 50% of the time for a static parameter approach.

A Multi-Frame GLMB Smoothing Based on the Image-Observation Sensor for Tracking Multiple Weak Targets Using Belief Propagation

The previous multi-frame version of the generalized labeled multi-Bernoulli model (MF-GLMB) only accounts for standard measurement models. It is not suitable for application in the detection and tracking of multiple weak targets (low signal-to-noise ratio) due to the measurement information loss. In this paper, we introduce a MF-GLMB model that formally incorporates a track-before-detect scheme for point targets using an image sensor model. Furthermore, a belief propagation algorithm is adopted to approximately calculate the marginal association probabilities of the multi-target posterior density. In this formulation, an MF-GLMB model based on the track-before-detect measurement model (MF-GLMB-TBD smoothing) enables multi-target posterior recursion for multi-target state estimation. By taking the entire history of the state estimation into account, MF-GLMB-TBD smoothing achieves superior performance in estimation precision compared with the corresponding GLMB-TBD filter. The simulation results demonstrate that the performance of the proposed algorithm is comparable to or better than that of the Gibbs sampler-based version.

Задача управления сенсорной сетью на основе рандомизированного и многоагентного подходов и ее приложения

Tracking multiple targets is a classic signal processing problem that occurs in many applications such as air, maritime and traffic control. Autonomous sensor networks serve as desirable platforms for multipurpose tracking due to their redundancy and reconfigurability. However, the network implementation makes it impossible to use the classical centralized approaches to filtering, since each sensor has limited computing power and limited access to the measurements of other sensors. In addition to topological limitations (each sensor can only communicate with several neighboring network nodes), communication between sensors can be limited, for example, due to limited bandwidth of communication channels, delay and data distortion. This article proposes a new algorithm for distributed tracking of multiple targets in a sensor network, which is a combination of the SPSA algorithm and the local voting protocol. The algorithm is consolidated under conditions of unknown but limited noise, the algorithm step size is optimized, and simulation is carried out to confirm the algorithm’s performance. Possible applications for the algorithm are also described.

Seamless Multiple-Target Tracking Method Across Overlapped Multiple Camera Views Using High-Speed Image Capture

Multiple-target tracking across multiple camera views is required for various practical tracking and surveillance applications. This study proposes a seamless tracking method based on label transfer using high-speed image capture. The proposed method does not require camera calibration to estimate the extrinsic parameters, and can be applied with simple line calibration. Therefore, it is effective for large-scale camera networks. In addition, owing to the features of the high-speed image capture and self-window, our method ensures seamless tracking of objects across multiple camera views. The feasibility of the method is proven through an experiment using a high-speed camera network system.

Three-Dimensional Multi-Target Tracking Using Dual-Orthogonal Baseline Interferometric Radar.

Multi-target tracking (MTT) generally needs either a Doppler radar network with spatially separated receivers or a single radar equipped with costly phased array antennas. However, Doppler radar networks have high computational complexity, attributed to the multiple receivers in the network. Moreover, array signal processing techniques for phased array radar also increase the computational burden on the processing unit. To resolve this issue, this paper investigates the problem of the detection and tracking of multiple targets in a three-dimensional (3D) Cartesian space based on range and 3D velocity measurements extracted from dual-orthogonal baseline interferometric radar. The contribution of this paper is twofold. First, a nonlinear 3D velocity measurement function, defining the relationship between the state of the target and 3D velocity measurements, is derived. Based on this measurement function, the design of the proposed algorithm includes the global nearest neighbor (GNN) technique for data association, an interacting multiple model estimator with a square-root cubature Kalman filter (IMM-SCKF) for state estimation, and a rule-based M/N logic for track management. Second, Monte Carlo simulation results for different multi-target scenarios are presented to demonstrate the performance of the algorithm in terms of track accuracy, computational complexity, and IMM mean model probabilities.

Survey of state estimation based on variational bayesian inference

State estimation problem in multiple target tracking usually faces high-dimensional uncertainty, including target model uncertainty, data association uncertainty, deep coupling and so on. Variational Bayesian inference provides a way to get the approximation for high-dimensional intractable problem. In this paper, we give the survey of state estimation based on variational Bayesian inference.

Multiple‐model Gaussian mixture probability hypothesis density filter based on jump Markov system with state‐dependent probabilities

The Gaussian mixture probability density (GM-PHD) filter has become a popular approach to solve the multiple-target tracking (MTT) problem because it can effectively and efficiently estimate the number of targets and target states that change over time from noisy measurements. In the GM-PHD filter, the target detection and survival probabilities, and the birth rate are assumed to be constant, irrespective of the target state. However, in some applications, for example, when the target follows the planned trajectory, the detection, survival and birth of the target depend on its state. Besides, when the target reaches a waypoint along the planned trajectory, it will take a manoeuvre to go to the next waypoint, but the GM-PHD filter does not accommodate manoeuvring targets that switch between several motion models. To address this, we propose a multiple-model GM-PHD filter with state-dependent probabilities, which can explicitly consider the mode transition probabilities, the probabilities of detection and survival, and the birth rate as a function of the target state. The performance of the proposed algorithm is demonstrated by illustrative MTT scenarios which include target manoeuvres during occlusion.

Distributed Soccer Training Smart Sensors for Multitarget Localization and Tracking

This paper presents an in-depth study and analysis of the localization and tracking of multiple targets in soccer training using a distributed intelligent sensor approach. An event-triggered mechanism is used to drive the acoustic array sensors in the distributed acoustic array sensor network, which solves the problem of increased communication load caused by frequent communication of microphones and effectively reduces the communication load between microphones as well as the energy consumption of the acoustic array sensor network. By designing a suitable state estimation equation for the acoustic source target and fully utilizing the measurement and state estimation information of its nodes as well as the state estimation information of neighboring nodes, the next moment state of the acoustic source target can be accurately predicted. A correlation filtering tracking algorithm based on multiscale spatial co-localization is proposed. In the proposed algorithm, the tracker contains a total of several subfilters with different sampling ranges. Then, this paper also proposes a collaborative discrimination method to judge the spatial response of the target image samples of each filter and jointly localize the target online. Based on this, this paper further explores the potential of correlation filter tracking algorithms in complex environments and proposes a robust correlation filter tracking algorithm that fuses multiscale spatial views. The cross-view geometric similarity measure based on multiframe pose information is proposed, and the matching effect is better than that based on single-frame cross-view geometric similarity; to solve the problem of player appearance similarity interference, a graph model-based cross-view appearance similarity measure learning method is further proposed, with players in each view as nodes, player appearance depth features as node attributes, and connections between cross-view players as edges to construct a cross-view player graph. The similarity obtained by the graph convolutional neural network training is better than the appearance similarity calculated based on simple cosine distance.

Tracking Multiple Targets Using Bearing-Only Measurements in Underwater Noisy Environments.

This article handles tracking multiple targets using bearing-only measurements in underwater noisy environments. For tracking multiple targets in underwater noisy environments, the Gaussian Mixture Probability Hypothesis Density (GM-PHD) filter provides good performance with its low computational load. Bearing-only measurements are passive and do not provide position information of a target. Note that the nonlinearity of the bearing-only measurements can be handled by Extended Kalman Filters (EKF) when applying the GM-PHD filter. However, range uncertainty of the target is large for bearing-only measurements. Thus, a single EKF leads to poor performance when it is applied in the GM-PHD. In this article, every bearing measurement gives birth to multiple target samples, which are distributed considering the feasible range of the passive sensor. Thereafter, every target sample is updated utilizing the measurement update step of the EKF. In this way, we run multiple EKFs associated to multiple target samples, instead of running a single EKF. To the best of our knowledge, our article is novel in tracking multiple targets in noisy environments, using the observer with bearing-only measurements. The effectiveness of the proposed GM-PHD is verified utilizing MATLAB simulations.

Safety-Critical Model-Free Control for Multi-Target Tracking of USVs with Collision Avoidance

Dear editor, This letter addresses the multi-target tracking of underactuated unmanned surface vehicles (USVs) subject to multiple stationary/moving obstacles. The kinetic model parameters of each USV are totally unknown. A safety-critical model-free control method is proposed for tracking multiple targets with a collision-free containment formation. Specifically, a distributed containment extended state observer (DCESO) is designed to estimate the convex hull spanned by the multiple targets. At the kinematic level, a collision-free kinematic guidance law is presented using a control barrier function (CBF) and an extended state observer for each follower USV. At the kinetic level, a model-free position tracking control law by using an adaptive ESO (AESO) is presented for each follower USV. By the designed safety-critical model-free control method, cooperative tracking of multiple targets under multiple stationary/moving obstacles can be achieved using completely unknown kinetic model parameters. Simulations are provided to illustrate the efficacy of the proposed safety-critical model-free control method for a fleet of USVs.

Modified Smoothing Algorithm for Tracking Multiple Maneuvering Targets in Clutter

This research work extends the fixed interval smoothing based on the joint integrated track splitting (FIsJITS) filter in the multi-maneuvering-targets (MMT) tracking environment. We contribute to tackling unknown dynamics of the multi-maneuvering-targets (MMT) using the standard kinematic model. This work is referred to as smoothing MMT using the JITS (MMT-sJITS). The existing FIsJITS algorithm is computationally more complex to solve for the MMT situation because it enumerates a substantial number of measurement-to-track assignments and calculates their posteriori probabilities globally. The MMT-sJITS updates a current target track by assuming the joint (common) measurements detected by neighbor tracks are modified clutters (or pretended spurious measurements). Thus, target measurement concealed by a joint measurement is optimally estimated based on measurement density of the modified clutter. This reduces computational complexity and provides improved tracking performance. The MMT-sJITS generates forward tracks and backward tracks using the measurements collected by a sensor such as a radar. The forward and backward multi-tracks state predictions are fused to obtain priori smoothing multi-track state prediction, as well as their component existence probabilities. This calculates the smoothing estimate required to compute the forward JITS state estimate, which reinforces the MMT tracking efficiently. Monte Carlo simulation is used to verify best false-track discrimination (FTD) analysis in comparison with existing multi-targets tracking algorithms.

Improved DeepSORT Algorithm Based on Multi-Feature Fusion

At present, the detection-based pedestrian multi-target tracking algorithm is widely used in artificial intelligence, unmanned driving cars, virtual reality and other fields, and has achieved good tracking results. The traditional DeepSORT algorithm mainly tracks multiple pedestrian targets continuously, and can keep the ID unchanged. The applicability and tracking accuracy of the algorithm need to be further improved during tracking. In order to improve the tracking accuracy of the DeepSORT method, we propose a novel algorithm by revising the IOU distance metric in the matching process and integrating Feature Pyramid Network (FPN) and multi-layer pedestrian appearance features. The improved algorithm is verified on the public MOT-16 dataset, and the tracking accuracy of the algorithm is improved by 4.1%.