Research On Obstacle Detection Of Wall Climbing Robots Based On Multi-sensor Data Fusion

This thesis aims at designing an obstacle detection method based on the information fusion of ultrasonics and camera. The main work consists of five parts: video acquiring, camera calibration, edge extraction from obstacle, range information fusion, comprehensive experiment.The paper firstly has introduced the main architecture and program flow of DirectShow, which was use to capture video to imagine through wireless camera; then we have illustrated camera model and the relation between each coordination system and maken use of Zhang Zhengyou planar template-based approach to calibration, by which the interior and exterior camera parameters has been determined. It lays a good foundation of image processing and obstacle position.To acquire more accurate obstacle information, this paper has presented a method which combines region with edge information: mean shift algorithm classified points contracted to the same pixel as one class, which implemented a multi-resolution segmentation, canny extracted edge, and then obstacle edges has been extracted from environments by filtering the image by length factors. The final result has proved this algorithm is efficient, especially when used in simple wall environment.Obtaining distance information has been based on an approach using geometric model of two ultrasonic transducers and extended discrete kalman filter(EDKF). This model has connected the measured distances from transducers and line parameters in ultrasonic coordination. This article has given the detail illustrations of EDKF parameter. The result of distance fusion is encouraging.Finally, this thesis has illustrated the hardware construction of this multisensor system and explained the key program flow. By using the edge information and distance information in inverse perspevtiv mapping, we has gotten the 3D coordinates in world coordination. Proved by comprehensive experiment, this system is reliable.