3D Trajectory Reconstruction Using Color-Based Optical Flow and Stereo Vision

Abstract

Automatic trajectory estimation of a moving object in a video is one of the most active research areas of computer vision, which finds many practical applications, such as development of sport playing robots, predicting trajectory for avoiding obstacle collision, automatic navigation of driverless vehicles, monitoring target hitting, etc. However, most of the work reported in literature only considers monocular videos. Due to the availability of low price stereo cameras, many applications take their advantages by incorporating depth information. In this paper, the 3D trajectory of a primary-color (red or green or blue) object is estimated using color-based optical flow and stereo vision. The purpose of using stereo vision is to gain depth information for generating 3D trajectory. The system has been tested on many stereo videos and experimental results are quite accurate. Besides, the low computation time required for finding depth of the tracked path makes it suitable for real time applications.