PBJDT: Point-Based Joint Detection-and-Tracking

Multiple object tracking is a more and more important technique with the rapid development of autonomous driving. In this work, we introduce a keypoint-based multiple object tracking method aim to address the challenges in complex road environments. It utilizes anchor-free keypoint detection to reduce computational resources while extracting robust features through an improved deep learning model architecture. We implement a background suppression technique to minimize false detections and incorporate the information from adjacent frames to capture object motion characteristics. To address occlusion and overexposure scenarios, we innovatively combine a Kalman filter with the detector and dynamically adjust the tracking strategy based on detection confidence. Experimental results demonstrate that the proposed method can be performed in real time and achieve a MOTA of 92.51% on the KITTI tracking dataset, which outperforms state-of-the-art methods.