Study On Autonomous Robot Vision Information Processing And Navigation

Autonomous mobile robot navigation is one of the most important research branches of intelligent robot. Vision-based navigation is the pivotal technology and research focus because of its overwhelming advantages than other type of sensors navigation. This thesis contains the following research based on professor Marr's vision theory. Firstly, robot low level image processing algorithms had been studied. Secondly, the vision system had been modeled. Thirdly, movement detection and vision-based navigation had been studied under the condition of dynamically real-time background.Low level image processing is the base of navigation. The paper discusses time domain filtering and frequency domain filtering according to the noise disturbance in the surroundings images. Then, classic edge detectors are introduced. Base on the analysis of wavelet edge detection algorithm, a type of self-adaptive wavelet edge detection is proposed to obtain much more clear and precise image edge.System structure of vertical iso-vision system with two CCD cameras is introduced in this paper. Mathematical modelling of robot vision navigation has been researched according to the camera perspective theory. The author discussed the mapping relation among the world coordinate, camera coordinate and image coordinate. Camera calibration is studied while the active vision-based camera self- calibration is introduced as well.The thesis totally introduces the whole structure of robot movement detection and navigation. Techniques such as self-adaptive background difference and Kalmann filtering and prediction are studied respectively. The former one applies difference algorithms between consecutive frames and dynamic background refreshing model to recognize moving objective while the later one combines invariant moment theory and Kalmann filtering together to realize movement detection and navigation. It is given to obtain both the position coordinate and relative velocity vector of the moving objective feature point effectively when Kalmann filtering algorithm is used.Corresponding algorithms are demonstrated by experiment using AS-R...