Region-Global feature fusion (RGF) algorithm for GM-APD range image and infrared image

The fusion of Geiger-mode avalanche photodiode(GM-APD) LiDAR and infrared detection can significantly enhance the far-distance detection and multi-target recognition capabilities. However, most existing fusion methods focus on infrared images and visible images. They are almost based on global images and cannot effectively utilize regional differences and consistency within regions to distinguish multiple targets. To address this problem, we propose a novel Region-Global Feature Fusion (RGF) algorithm that focuses on the GM-APD and infrared images, which could enhance multi-target classification capability by combining global and regional feature fusion methods. In the proposed algorithm, we correct the super-pixel segmentation result by evaluating the region complexity and propose a dual regularization joint constraint (IN-TV) to balance the global maps and regional maps for maximizing their advantages. The AA and Kappa of the proposed method are 155% and 178% higher than the infrared image, and also higher than the other two state-of-the-art fusion algorithms. This work provides a robust technical foundation for all-weather, multi-target detection systems.