Tracking a single object in dynamic point cloudDynamic point cloud sequencesPoint cloud sequences is a crucial task for various intelligent systems. In this chapter, we propose DeepPCT, a deep neural network designed for precise and robust object trackingObject tracking in sequential point clouds. Unlike conventional methods that rely on bounding box classification and regression for localization, our framework directly optimizes Intersection-over-Union (IoU) prediction to improve 3D localization accuracyLocalization accuracy. To overcome the limitations of the standard IoU metric, we introduce a central distance normalized IoU, which effectively reduces ambiguity in object localization. In addition, we explore the benefits of incorporating a re-detection module to improve tracking robustness, as failures often arise from error accumulation and significant variations. Extensive experiments on the KITTI benchmark show that our tracker attains strong performance, with a 3D Precision of 63.7% and a 3D Success of 45.0%. Furthermore, our method demonstrates robust performance in challenging scenarios with severe illumination changes and occlusions, even outperforming existing 2D state-of-the-art trackers.