Although infrared small target detection has been broadly used in airborne early warning, infrared guidance, surveillance and tracking, it is still an open issue due to the low signal-to-noise ratio, less texture information, background clutters, and so on. Aiming to detect a small target in an infrared image with complex background clutters, this paper carefully studies the characteristics of a target in an IR image filtered by the difference of Gaussian filter, concluding that the intensity of the adjacent region around a small infrared target roughly has a Mexican-hat distribution. Based on such a conclusion, a raw infrared image is sequentially processed with the modified top-hat transformation and the difference of Gaussian filter. Then, the adjacent region around each pixel in the processed image is radially divided into three sub-regions. Next, the pixels that distribute as the Mexican-hat are determined as the candidates of targets. Finally, a real small target is segmented out by locating the pixel with the maximum intensity. Our experimental results on both real-world and synthetic infrared images show that the proposed method is so effective in enhancing small targets that target detection gets very easy. Our method achieves true detection rates of 0.9900 and 0.9688 for sequence 1 and sequence 2, respectively, and the false detection rates of 0.0100 and 0 for those two sequences, which are superior over both conventional detectors and state-of-the-art detectors. Moreover, our method runs at 1.8527 and 0.8690 s per frame for sequence 1 and sequence 2, respectively, which is faster than RLCM, LIG, Max–Median, Max–Mean.

The second association enhances Multi-Object Tracking (MOT) by reducing missed detections and trajectory fragmentation but is limited by the poor distinguishability of low-confidence detections. To address this, we introduce interactive features leveraging Graph Neural Networks (GNNs) to enhance object distinction. Unlike existing GNN-based trackers that compute interactive features for all objects, we selectively calculate interactive features based on directed reliable neighbor graphs for objects in the second stage. These graphs include two kinds of nodes: reliable nodes (already associated in the first association) and unreliable nodes (remaining objects). Bidirectional edges between reliable nodes indicate matches, while directed edges from reliable to unreliable nodes represent neighbor relationships. These graphs are forwarded to graph attention networks to obtain interactive features combined with motion features for the second association. Experimental results on the MOT17 (65.3 in HOTA, 80.6 in IDF1) and MOT20 (63.8 in HOTA, 78.0 in IDF1) benchmark datasets demonstrate the effectiveness of our proposed tracker, particularly in HOTA and IDF1.