Research on deep learning technology to analyze the behavior of multi-target objects

Abstract

In multi-target tracking technology, the application of deep learning technology can effectively improve the accuracy of target detection, but because the target movement is irregular, the shooting Angle and viewpoint will change, and there is occlusion between the targets, which will affect the final detection results. Therefore, after understanding the theory of multi-target tracking technology with detection and tracking as the core, researchers focus on how to avoid target switching caused by false detection or missing detection, and associate multiple target detection information in the video with historical trajectories. On the basis of understanding the research status of multi-target tracking technology, this paper proposes a multi-target detection algorithm and association method with deep learning as the core. The final experimental results show that the improved network has a positive impact on the accuracy of multi-target detection and can fully meet the requirements of target tracking processing in complex environment.