Lightweight RGBT Target Tracking Algorithm that Incorporates an Attention Mechanism

A lightweight Siamese network-based target tracking enhancement technique is developed in response to the issue that the current target tracking algorithms, which combine the two modalities of visible light and thermal infrared, are unable to adequately account for real-time and resilience. A lightweight backbone network is used to extract the features of the two modalities, with a focus on the crucial features of position and channel information. Thermal infrared modal features are also added, which significantly improves the target's semantic information. The algorithm's tracking performance is further enhanced by utilizing the dual-stream Siamese network's multi-branch inter-correlation method. In the public datasets GTOT and RGBT234, the algorithm presented in this research achieves 88.6%/71.4% and 82.3%/58.5%, respectively, with an operating speed of 73.8fps. The experimental results demonstrate that the approach in this study performs better in low light, temperature crossover, and deformation scenarios, and has improved real-time performance and robustness.