Monocular-vision-based study on moving object detection and tracking

In this paper, we present an approach for recognizing and tracking a moving color object which break into the robot's camera vision in the mobile robot. In order to obtain input feature for moving-object recognition, we perform image processing of the video sequence by background substraction method. Given the images of video sequence provided by a fixed camera on mobile robot, disparity images in different time are integrated into a tracking framework in order to find the HSV space based color features of the moving object. Based on the color object's moving and background substraction method, an color classifier based on the HS thresholds is trained to detect moving object. We show experimentally that the moving-object recognition performance can be improved significantly by using information about object's edge as an additional feature. Through the recognition, we can find the center of the moving object and control the mobile robot's angle and rate to realize the vision-tracking of moving object. Our system aims at applications in the field of robot intelligence,where it is important to do run-on recognition in real-time, to allow for robot sel-learning and not to rely on appointed features.