The Optimization of Object Detection and Localization in Complex Background for Vision-based Robot

摘要

This paper studies the object detection and localization for vision-based robots. The vision-based robot collects the stream of images by using the wide-angle monocular camera to get more-detailed information of objects around the robot. Based on the captured camera image, we first preprocess the collected image information to extract useful points by distortion correction, color space conversion and noise reduction. After getting color and shape information of object, we propose several algorithms to cancel the interfere from complex background with a close operation, to use a multi-threshold limitation for object detection and localization, and to find minimum circumscribed rectangle of target in order to recognize obstacles and adjust robot's direction and position. Finally, the robot performs well with the presented methods.