Performance Analysis of Siamese Tracker Through Preprocessing of Infrared Images

Target tracking in infrared images is actively researched in computer vision. Unlike RGB images, infrared images rely on a single channel of thermal signals, making them vulnerable to issues such as non-uniform illumination and noise. In this study, we propose integrating CLAHE (Contrast-Limited Adaptive Histogram Equalization), a preprocessing technique, into the SiamRPN framework, a widely used Siamese network-based tracker. CLAHE enhances local contrast while maintaining computational efficiency, enabling robust tracking, even in resource-constrained environments. The performance of the proposed method was evaluated using challenging subsets of the LSOTB-TIR dataset, selected to represent conditions where standard models typically struggle. Experimental results demonstrate that the CLAHE-enhanced SiamRPN achieves significant improvements in accuracy and robustness compared to the baseline model, particularly in environments with low contrast and complex backgrounds. These findings suggest that CLAHE can be a practical solution for improving performance in environments with limited computational resources.