Deep Learning and Depth Integrated Method for Visual Tracking of Object Under Complicated Background

Visual tracking of objects in complex environment is generally of low accuracy and high identity switching rate due to the changes of visual and motion characteristics, which brings challenges to the complete prediction of the object trajectory and can hardly be overcome by traditional object tracking algorithms. This paper introduces a novel object tracking model which enhances tracking accuracy by integrating deep learning and depth. The tracking algorithm centered on the acquisition of depth can not only facilitate the retrieval of lost objects but also expeditiously eliminate falsely detected objects. This approach effectively reduces interruptions in object tracking trajectories during object motion. Rigorous tests in diverse scenarios show the proposed algorithm achieves a tracking accuracy (MOTA) of 88.12%, representing a substantial enhancement of 3.54% over the baseline algorithm, and witnessing a noteworthy 73.68% reduction in identity switch frequency.