Research on Tracking and Localization Algorithm of Motion Target Based on Binocular Stereo Vision

Machine vision, as an important branch of computer science, has developed rapidly over the past three decades. Binocular stereo vision is an important part of machine vision, and binocular stereo vision can simulate the human eye to provide image 3D position information, but binocular stereo matching is often limited by the environment and can not be accurate 3D reconstruction, and at the same time, the tracker of the moving target in the complex environment has poor robustness. This paper proposes a real-time tracking and localization method for moving objects in 3D space based on a binocular vision information system. Through the optical imaging of the binocular vision system and the improved twin neural network target tracking algorithm, the object-point mapping relationship of the provided mathematical model and the trajectory of the center mass point on the sequence of visual images are obtained. In addition, the outdoor binocular vision-based motion target tracking and localization experimental surface compared with the traditional tracking and localization method, the tracking time of this method is significantly shortened to 90 fps, and the maximum relative error rate between the motion depth value and the actual value of the localization of this method is not more than 0.9, respectively, which verifies the validity and accuracy of the model in this paper.