Vision Based Moving Target Tracking Of Mobile Robots

With the continuous development of robot technology, mobile robots find more and more application fields, and recently vision based navigation becomes one of the hottest research areas in mobile robots.This study mainly focuses on moving target tracking technology for mobile robots based on monocular vision. From the practical point of view, a moving target tracking scheme is introduced for a mobile robot. The scheme is designed with a modular method, and comprises of system starting block, image processing block and motion control block. Each block has its own function, and simultaneously realizes image segmentation, object recognition, object localization and tracking as a whole.In the image processing block, a processing algorithm is proposed to meet the needs of real time and precision. Firstly, a Kalman filter is used to predict the position and the velocity of a tracked moving object, and then the region of interest in the video frame is selected based on the predicted position and velocity information. In the image segmentation, a median filter is used to process the gray images, and then a threshold method is introduced to segment an image into different regions of interests, and the regions are further segmented into independent candidates of the tracked object with mathematical morphological processing. Finally, the features of the candidates are extracted and used to recognize the tracked moving object.In the motion control block, a calibration algorithm based on Zhang's method is first introduced and used to obtain a camera parameters, and then the relationship between the camera coordinate system, the image coordinate system and the world coordinate system is established using the pinhole camera model, and then compute the location of the tracked mobile object in the 3d space. Finally, a control scheme is proposed to track the moving target.All the algorithms are realized by Visual C++, and proved by the experiment results that a moving object is able to be recognized precisely, and tracked by a mobile robot system in real time.