Data Association for Multi-Object Tracking Using Assignment Algorithms

Multi-object tracking (MOT) plays a very important role in advanced applications such as driver assistance systems, autonomous vehicles, aircraft monitoring systems. One of the fundamental problems of multi-object tracking is the unknown data association. At each time step, we get multiple sensor measurements. However, we don't know where these sensor measurements come from. Each measurement can come from one of the objects detected in the previous time step, from a new object entering the field of view (FOV), or from clutter. As the number of objects and measurements increases, the combinatorial nature causes the number of data association hypotheses to increase rapidly, making it impossible to consider all data associations [1,2]. One approach to solving the data association problem is to transform it into an optimal assignment problem. One approach to solving the data association problem is to use algorithms of the optimal assignment problem. In this paper, we will introduce the method of using optimal assignment algorithms to solve the data association problem in MOT. At the same time, we also compare the performance of these algorithms in terms of data association accuracy and computational complexity. The results showed that using assignment algorithms like Gibbs sampling brings high performance in data association.