Infrastructure-supported Cooperative Tracking Using Gaussian Mixture Probability Hypothesis Density

In recent years, targets tracking plays an important role in the rapidly expanding field of autonomous driving. With the presentation of Vehicle-to-Everything (V2X) communication standards and Mobile Edge Computing (MEC) technology, the cooperative tracking method that relies on roadside infrastructure can provide connected traffic participants with higher-precision information. However, in the actual intersection scenes, small targets such as pedestrians are at close range and are easy to be occluded for a long time in the intersection scenes. In this paper, we propose a novel framework for cooperative tracking of radars and cameras, and focuses on a modified method of Gaussian Mixture Probability Hypothesis Density Filter (GM-PHD) for occluded objects, and finally, we evaluate the performance of the algorithm on a real intelligent intersection platform. The result shows that the modified GM-PHD improves tracking performance by 69.6% compared with the traditional association-based method, and by 70.5% compared with the original GM-PHD. Consequently, the proposed method can effectively enhance the security and reliability of autonomous driving assistance systems, and its application has potential to support the construction of smart cities.