A study of 3D feature tracking and localization using a stereo vision system

With the advances of image processing and computer vision, techniques being developed are no longer limited in images acquired with single camera (namely plane vision). A stereo vision system with two cameras has become the research of interest in many areas because its ability to yield the depth information similar to human vision. In this study, the objective was to develop a system that can automatically track and localize a 3D feature in motion using left and right video sequences. The system design included feature definition, feature tracking, feature localization, and depth computation. In addition, we evaluated our system with several research parameters, including various depth s, video tracking in plane vision or stereo vision, use of kernel functions, and integer-pixel vs. sub-pixel accuracy. Our results demonstrated that the system could track and localize the given feature in motion, leading to yield reasonable results of depth information. In addition, the video tracking in stereo vision with sub-pixel accuracy clearly outperformed the video tracking in plane vision with integer-pixel accuracy. In summary, our system yielded a potential solution in tracking and localizing 3D feature that could be incorporated in a large variety of video applications.