A Novel Multiple Object Tracking Algorithm for Autonomous Vehicles

摘要

Multiple object tracking is a vital task for autonomous vehicle environment perception. In this paper, we design a novel multi-object tracking method for autonomous vehicles. In the detection section, we utilize popular Faster-RCNN as our baseline method. Then, in data association, we combine appearance, motion, and interaction model to build a unified feature descriptor to explore the nature of tracking object. We evaluate our algorithm on a popular and standard benchmark and compare with the state-of-the-art methods. The results denote that our algorithm achieve good performance at high frame rates.