Comparative study of data association approaches for multiple vehicle tracking

Abstract

The primary objective of this paper is to make a comparative study of different data association methods for multiple target tracking (MTT) problem, analyze the problems encountered in almost all target tracking applications and propose a new method, considering the comparison results of performed algorithms. As the first solution to the MTT problem, the Global Nearest Neighborhood (GNN) approach is implemented. Although it is the easiest to implement, its drawbacks in target tracking are observed, which justified the investigation of probabilistic approaches like the probabilistic data association filter (PDAF) and the joint probabilistic data association filter (JPDAF). In this paper, those three methods are compared in terms of different error criteria and results are presented. Considering the implemented metrics, the results of JPDAF are observed to be most efficient among them by its ability to track multiple targets in a cluttered environment. Thus, a new method is developed by improving JPDAF, with the aim of eliminating its weaknesses, which is observed when the target number changes during tracking. Improved method consists of an additional kill and initialize track algorithm, by which the first assumption of JPDAF about the known target number in clutter, will not cause any problem when the targets leave or enter the environment randomly during tracking. Along with that modification, new algorithm is developed considering the flexibility in maximum allowable target number and code generation for real-time applications. Functionality of developed algorithm is independent of the number of sensor measurement and tracked objects.