Track-oriented Multiple Hypothesis Tracking Research Articles

Multi-appearance segmentation and extended 0-1 programming for dense small object tracking.

Aiming to address dense small object tracking, we propose an image-to-trajectory framework including tracking and detection, where Track-Oriented Multiple Hypothesis Tracking(TOMHT) is revised for tracking. Unlike common cases of multi-object tracking, merged detections and the greater number of objects make dense small object tracking a more challenging problem. Firstly, we handle frequent merged detections through the aspects of detection and hypothesis selection. To tackle merged detection, we revise Local Contrast Method(LCM) and propose a multi-appearance variant, which exploits tree-like topological information and realizes one threshold for one object. Meanwhile, one-to-many constraint is employed via the proposed extended 0-1 programming, which enables hypothesis selection to handle track exclusions caused by merged detections. Secondly, to alleviate the high complexity caused by dense objects, we consider batch optimization and more rigorous and precise pruning technologies. Specifically, we propose autocorrelation based motion score test and two-stage hypotheses pruning. Experimental results are presented to verify the strength of our methods, which indicates speed and performance advantages of our tracker.

Multiple-Hypothesis Tracking for Targets Producing Multiple Measurements

This paper extends the multiple-hypothesis tracking (MHT) recursion to allow for multiple detections per target, per scan of data. We show that the generalized recursion admits the factorization of global-hypothesis probabilities, enabling track-oriented MHT. We propose a two-stage MHT implementation that allows tractable processing in this setting, and explore performance with respect to classical track-oriented MHT processing.

Track-Oriented Multiple Hypothesis Tracking Based on Tabu Search and Gibbs Sampling

In order to circumvent the curse of dimensionality in multiple hypothesis tracking data association, this paper proposes two efficient implementation algorithms using Tabu search and Gibbs sampling. As the first step, we formulate the problem of generating the best global hypothesis in multiple hypothesis tracking as the problem of finding a maximum weighted independent set of a weighted undirected graph. Then, the metaheuristic Tabu search with two basic movements is designed to find the global optimal solution of the problem formulated. To improve the computational efficiency, this paper also develops a sampling based algorithm based on Gibbs sampling. The problem formulated for the Tabu search-based algorithm is reformulated as a maximum product problem to enable the implementation of Gibbs sampling. The detailed algorithm is then designed and the convergence is also theoretically analyzed. The performance of the two algorithms proposed are verified through numerical simulations and compared with that of a mainstream multiple dimensional assignment implementation algorithm. The simulation results confirm that the proposed algorithms significantly improve the computational efficiency while maintaining or even enhancing the tracking performance.

A Batch Algorithm for Estimating Trajectories of Point Targets Using Expectation Maximization

In this paper, we propose a strategy that is based on expectation maximization for tracking multiple point targets. The algorithm is similar to probabilistic multi-hypothesis tracking (PMHT) but does not relax the point target model assumptions. According to the point target models, a target can generate at most one measurement, and a measurement is generated by at most one target. With this model assumption, we show that the proposed algorithm can be implemented as iterations of Rauch-Tung-Striebel (RTS) smoothing for state estimation, and the loopy belief propagation method for marginal data association probabilities calculation. Using example illustrations with tracks, we compare the proposed algorithm with PMHT and joint probabilistic data association (JPDA) and show that PMHT and JPDA exhibit coalescence when there are closely moving targets whereas the proposed algorithm does not. Furthermore, extensive simulations c comparing the mean optimal subpattern assignment (MOSPA) performance of the algorithm for different scenarios averaged over several Monte Carlo iterations show that the proposed algorithm performs better than JPDA and PMHT. We also compare it to benchmarking algorithm: $N$ -scan pruning based track-oriented multiple hypothesis tracking (TOMHT). The proposed algorithm shows a good tradeoff between computational complexity and the MOSPA performance.

If a tree falls in the woods, it does make a sound: multiple-hypothesis tracking with undetected target births

This paper introduces a generalization of the multiple-hypothesis tracking (MHT) formalism for multitarget tracking. To our knowledge, MHT treatments in the literature do not consider undetected target birth events. Their inclusion leads to an interesting extension to the MHT recursion and necessitates aggregation over indistinguishable global hypotheses. We show that the MHT recursion factors, enabling track-oriented MHT (TO-MHT), albeit with clusters of indistinguishable undetected births. The treatment requires a distinction between those targets that are eventually detected (we call these unnoticed targets) and those that are never detected (we call these ghost targets). While the formulation appears more complex, there is structure to the solution that can be exploited, resulting in the same number of relevant track hypotheses for detected targets as in the classical TO-MHT solution. In the time-invariant case, the solution simplifies further because we need not consider unnoticed targets and there is a fixed structure to the ghost target solution.

An Efficient MHT Implementation Using GRASP

The multiple hypothesis tracking (MHT) approach has been proven to be successful in multiple target tracking applications, however, its computational complexity remains a major hurdle to its practical implementation. This paper presents an efficient MHT implementation, referred to as “GRASP-MHT”, which integrates a greedy randomized adaptive search procedure (GRASP) within a track-oriented MHT framework. The hypothesis generating problem arising in the MHT framework is formulated as a maximum weighted independent set problem, and a GRASP module is designed to generate multiple high-quality hypotheses. An extensive simulation study was carried out to compare the performance of the proposed GRASP-MHT against several well-known multitarget tracking algorithms, and multiple metrics were considered in order to make the performance evaluation more comprehensive. Experimental results indicate that, by efficiently generating and fusing multiple high-quality global hypotheses in the data association process, GRASP-MHT is able to achieve better overall tracking performance than other algorithms, especially in a closely-spaced multitarget scenario.