Vision-Based Navigation And Object Detection System For Mobile Manipulator

Vision navigation is being a developing trend of mobile manipulator's navigation recent years. How to recognize man-made road and road sign rapidly and exactly by using vision technology and How to pick-up object figures are crucial problems in the study of mobile robot's vision navigation. Supported by the National Hi-Tech R&D Program (863) of China (Grant No. 2003AA421020).In this paper, a vision-based navigation and object detection system is provided, The main work and innovative ideas include:The vision system of HEBUT- II mobile manipulator is designed. The modeling of camera and the conversion between image coordinate and vehicle coordinate is completed, which makes the road information revert from the image space. Experiments on image segmentation are done and a method of image segmentation based on HSI color threshold is used for mobile manipulator's road and road-signs image segmentation. The image segmentation method based on HSI color threshold has many advantages: strong antinoise ability, using more information during image processing, perfect segmentation result without being affected by illuminative condition, and simple calculating process. All these advantages improve the segmentation speed and veracity of complicated road image.The mobile manipulator's road-signs recognition system is designed. Man-made guiding road and road sign are laid on the ground, which simplifies environment. The road and road sign have two different colors, and two different HSI color threshold coming from offline experiment are used for road image segmentation. Therefore, the complicated processes of determining threshold and distinguishing the road and road sign area are avoided and the real-time capability is improved. The recognition of road-sign is by artificial neutral net.The research on manipulator's orientation based on vision is carried on. It emphasizes on image-preprocessing, including: gray treatment, the choice of threshold, binarization, erosion, dilation, group color up method. These transformation can produce clear image, and the computer can recognize the image easily. The recognition of object figure is by invariant moments. The experiments prove this method can segment object exactly, so the leaking location is oriented.