High-precision and real-time visual tracking algorithm based on the Siamese network for autonomous driving

Visual object tracking is often used to track obstacles in autonomous driving tasks. It requires real-time performance while dealing with target deformation and illumination changes. To solve the above problems, this paper proposes a high-precision and real-time visual tracking algorithm for autonomous driving based on the Siamese network. First, our tracker utilizes ensemble learning to fuse two feature extraction branches that are derived from the convolutional neural network. Then, the channel attention mechanism is added before concatenation to redistribute feature weights. Finally, a region proposal network is adopted to generate tracking bounding boxes. Extensive experiments demonstrate that compared with the state-of-the-art algorithms, the proposed method achieves satisfactory results on four benchmark datasets while maintaining a higher frame rate. Also, the qualitative analysis results on the KITTI dataset indicate that our method can meet the challenges in autonomous driving.