Biological Eagle-Eye-Based Correlation Filter Learning for Fast UAV Tracking

As a cutting-edge tracking algorithm for unmanned aerial vehicles (UAVs), discriminative correlation filter (DCF)-based trackers have evolved by embracing diverse feature extraction and computational methods. However, many DCF-based trackers struggle to adapt to background variations encountered during UAV flights due to the lack of image feature enhancement and velocity-dependent scale adjustment. The high-speed movement of UAVs requires trackers that can leverage fast computational capabilities to keep pace. Taking inspiration from the impressive hunting skills of eagles, this article introduces a DCF-based tracker called eagle-eye-based correlation filter (EECF). The EECF tracker utilizes a 'dual fovea' mechanism for image preprocessing, enhancing the utilization of target information in candidate regions. By integrating color opponent and contrast adaptation into the feature extraction process, the EECF tracker emulates the perceptual characteristics of eagle vision, enabling robust target recognition in complex backgrounds. We introduce masking and regularization techniques based on the 'lateral inhibition' mechanism observed in eagle visual pathways. The EECF tracker showcases exceptional computational speed and surpasses other trackers in tracking success rates. Extensive experiments carried out on four publicly available UAV tracking benchmarks and custom drone datasets confirm the superiority of our tracker over other DCF-based options. Our algorithm achieves a maximum tracking speed of 86 FPS while maintaining higher accuracy than most comparative algorithms.