Research On Detection And Tracking Of Moving Obstacles Around Unmanned Vehicle Using Binocular Stereo Vision

Unmanned vehicle is an intelligent vehicle which has the capability of identifying and sensing surrounding environment by sensors, identifying obstacles automatically and controlling driving path autonomously. It has great advantages in terms of reducing traffic accidents, violations and labor intensity of the driver and the allocation of traffic resource for unmanned vehicles. Researches on the active identifying surroundings and monitoring obstacles techniques of unmanned vehicles are the key point of exerting the advantages of an unmanned vehicle in the transport system. Environment recognition technology as an important research method which based on binocular stereo vision has become the focus of the current research. In this paper, the method of detection and tracking of moving obstacles around unmanned vehicle using binocular stereo vision was studied. The main research contents are:(1)Obstacle position measurement model based on binocular stereo vision was established, which is suitable for unmanned vehicles. Pinhole camera model containing distortion coefficients and binocular camera system model composed by two cameras were built firstly. The obstacle position measurement model combined with an unmanned vehicle coordinate system and the assembly relation of binocular cameras on unmanned vehicle was built next.(2) Binocular stereo vision measurement test platform was built and calibrated. The platform consisted of two USB interface industrial camera and mounting bracket, and was calibrated with the classic Zhang plane calibration method. Internal and external camera parameters matrix was got after plane calibration. Rotation and translation matrix of the left and right camera position was also obtained by stereo calibration. According to the parameters and measurement model established in the last chapter,three-dimensional depth information matrix of the specific obstacle was calculated(3)CamShift obstacle tracking algorithm based on the combination of the depth information and histogram backprojection was proposed. Improved obstacle tracking algorithm was proposed and studied based on the depth information which came from the binocular stereo vision measurement test platform. The new proposed algorithm can solve the defect of traditional CamShift tracking algorithm, which needs to manually select the initial search locations and will lost track window when the hue of obstacle and the background are similar.(4)The improved tracking algorithm was proposed and the accuracy of the binocular stereo vision measurement test platform in this paper was validated. After measuring a particular object at a specific location and tracking moving obstacle in the particular scene in real time, the measurement accuracy of the platform and the feasibility of the improved obstacle tracking algorithm was validated.The research ideas of this paper are establishing the obstacle position measurement model, improving the traditional tracking algorithm and validating the measurement accuracy of the platform by experimental verification. Effective obstacle positioning calculation model and tracking algorithm based on depth information are proposed and studied, which are of great theoretical and practical values.