Research On Wall-climbing Robot’s Technique For Environment Detection Based On Binocular Stereo Vision

To ensure safe and reliable operation, the sensors are in need for wall-climbing robot to perceive the unknown environment. In recent years, with the development of image sensor technology, embedded technology, image processing technology, machine vision plays a important role in robot’s environment detection. According to the number of image sensors used, machine vision can be classified as monocular vision, binocular vision or multi-vision. As the binocular stereo vision technology has a huge advantage in three-dimensional measurement, it has become a hot topic.Signpost is global characteristics in the environment, which can be used as robot navigation and location reference. Obstacle was something blocking the robot forward in the movement direction, which is an important part of the environment. We focused on three-dimensional positioning methods of colored signs in scene and robot obstacle detection methods based on binocular stereo vision in this paper. The chief aspects are studied as follows:(1) Overall plan of binocular vision-based system for wall-climbing robot was designed, and the hardware and software platform binocular vision-based system was set up.(2) We calibrated binocular vision system for wall-climbing robot. To begin with, zhang’s method was used for monocular camera calibration, thus, the intrinsic parameters within each camera was caculated. In the same way, the relative position of the two cameras was obtained by stereo calibration. Then, Bouguet algorithm was employed to rectify original image pair so that pixel rows between the cameras were exactly aligned with each other. Accordingly, re-projection matrix for three-dimensional reconstruction was obtained.(3) Positioning methods of colored signs based on binocular stereo vision technology was studied in this paper. In the first place, smooth operation was used to reduce the image noise. Then, image threshold operations was performed to get the target pixels of colored signs in HSI color space.Then, matching features ware obtained by extracting the centroid of target pixels of colored signs. In addition, the position of colored signs was caculated by three-dimensional reconstruction. Finally, positioning error was calculated and analyzed.(4) Obstacle detection methods based on binocular stereo vision was studied in this paper. Above all, edge-detection algorithm was used to detect edge pixels of stereo image pair. At the same time, the location of edge pixels in particular direction was recorded. Then, feature vectors of edge pixels ware extracted by SURF descriptor and the nearest neighbor feature matching algorithm was used to find the matching features. At last, three-dimensional coordinates of the edge points was obtained. If the height of edge pixels in space was greater than a certain predetermined value, some of them ware recorded as obstacle points.