Research On Some Issues Of Visual Navigation Of Ground Intelligent Robot

Ground intelligent robot is a mobile robot, which can run autonomously andcontinuously in real-time indoor or outdoor, its research has became the tactic researchobject of high technology of all countries. With the promotion of urgent demand onmilitary, civilian and research fields, ground intelligent robot are changing with eachpassing day. Visual navigation is one of the core technologies of ground intelligent robot,so this paper focuses on some issues of it, mainly including visual simulation system inoutdoor nature environment, distortion correction, camera calibration, vision based mobilerobot stair following and stereo correspondence.With the development of research on ground intelligent robot, the capacity the robotwill grow quickly, which causes the price of sensors much higher, so the pre-research ofground intelligent robot is often done through simulation system. Based on that reason, webuild a visual simulation system of outdoor nature environment, which simulates CCDcolor camera, laser radar, and stereo vision. CCD camera is simulated through setting theparameters of view volume in Open GL, intercepting and saving the pixels of the screen.Ray-tracing method is used to simulate the Laser Radar, and Stereo Vision is simulated byacquiring the data from the scene. The system is able to simulate the acquired data ofsensors truly when the robot is moving. Using the data provided by the system, thealgorithms of fusion and plan modules is able to be designed.According to the feature of stairs in city environment, this paper proposed a fast andstable algorithm, which uses camera calibration parameters to rectify images, and fits allthe useful edges of detected images using least square; stair climbing modal was built torecover the offset angle of mobile robot, and at the end a control strategy was presented.Experiments in the real world environment both show that the algorithm is effective and ofgreat value to use.To resolve the distortion of cameras, this paper firstly proposes a straight line featurebased distortion correction method, and then do camera calibration using corrected images.The distortion correction method extracts many parallel lines of different direction fromdifferent images, and uses nonlinear least square to optimize the distortion parameters. Thethought of step by step is added when the optimization method working. 3D worldcoordination is not need to know, and the method is easy to implement and can be executed while robot moving. When doing camera calibration, corrected images are used to detectcomers, and a strategy is proposed to autonomously match the detected pixels with sensepoints.Dynamic programming is an often used stereo matching algorithm. To solve the wrongmatching problem in depth discontinuities of dynamic programming, this paper proposedan algorithm which can remove the wrong matches through left-right consistency, and usea fast method to fill in the blank. Because of the streaks in one scanline, and roughnessbetween different scanlines, ground control points based iterative dynamic programmingstereo matching algorithm is proposed, which can solve the two problems commendably.Algorithm based on global energy minimization is one of the best categories of stereomatching algorithm. We propose a segment based two-step algorithm, which firstlysegment the images into many sub-area, then fill in some areas with the most reliabledisparities, and at last using genetic algorithm to optimize the left disparities. Color andspace distance based belief propagation stereo matching algorithm is proposed, in whichcolor and space information of color images is proposed to build the disparity space image,the truncated linear model is used to define the smooth term of disparity, and at last beliefprorogation is used to optimize the full disparity image.