ECO-SWGST: ECO-based tracking with side window generalized structure tensor for real-time infrared small target tracking

Abstract

Infrared small target tracking is a crucial research direction in infrared search and track systems, and enhancing detection and tracking accuracy in complex backgrounds while meeting real-time requirements remains a challenging task. To enhance the performance of traditional correlation filter algorithms, we propose a side window generalized structure tensor (SWGST) detection framework for real-time infrared small target tracking. The proposed method is built upon the correlation filter framework efficient convolution operators (ECO) and integrates SWGST. SWGST uses an anisotropic neighborhood selection strategy and incorporates eight-direction window filtering during the computation of the generalized structure tensor. Extensive experiments demonstrate that the proposed model successfully combines the detection algorithm with the correlation filter tracking framework, significantly improving the accuracy and effectiveness of real-time tracking. This approach enhances the certainty map of small targets, significantly improving the discrimination between targets and background edges while maintaining computational efficiency.