Thermal Infrared Target Tracking Algorithm Based on KL Divergence and Channel Selection

To solve the problem that a single tracker cannot effectively deal with the complex background and significant changes in target appearance, leading to the problem of low accuracy of thermal infrared target tracking, a tracking algorithm based on a fully-convolutional Siamese network is proposed for thermal infrared tracking. First, a pre-trained convolution neural network is used to extract the features of multiple convolution layers of thermal infrared targets and select channels. On this basis, three corresponding trackers are constructed, and each tracker performs tracking independently and returns a response map. Then, the Kullback Leibler (KL) divergence is used to optimize and integrate multiple response maps to obtain a stronger response map. Finally, the integrated response map is used to determine the target location. To evaluate the performance of the proposed algorithm, experiments were conducted using the most comprehensive thermal infrared tracking benchmark, LSOTB-TIR. The experimental results show that the proposed algorithm can adapt to complex and diverse infrared tracking scenes, and its comprehensive performance is better than that of existing infrared tracking algorithms.