Abstract: Three-dimensional (3-D) tracking in point cloud is a core competence of autonomous robots to perceive and forecast the environment. How to initialize bounding box seeds and optimize their ...