Multi-3D-Object Tracking by Fusing RGB and 3D-LiDAR Data

Multiple object tracking (MOT) is a fundamental problem in the autonomous driving research community. Through accurate and efficient tracking, ego-vehicle can get the location velocity of surrounding objects and make a reasonable future motion planning. Different from most of the methods adopting the RGB or 3D-LiDAR data independently, this paper aims to track the perceived objects by fusing RGB and 3D-LiDAR data, the standard sensor configuration in current autonomous vehicles. Specifically, we firstly use Hungarian algorithm as a backbone model to associate the 3D point cloud of each object in adjacent frames. Then, we fully explore the appearance feature in RGB frame and geometrical feature in 3D point cloud to restrict the wrongly associate target IDs because of the interaction of near objects. We evaluate our method on the newly proposed BLVD dataset, and show the favorable performance.