Environment Mapping from Motion and Stereo Image Sequences

• Main Authors: ChungPing Hsia, 夏仲平
• Other Authors: Hsien-Huang Wu
• Format: Others
• Language: zh-TW
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/49970419419141688865

碩士 === 國立雲林科技大學 === 電機工程技術研究所 === 88 === In this thesis, we will develop a binocular stereo vision system which is to be set in an autonomous mobile robot (Autonomous Navigation Vehicle), assists to safety navigation indoors. The stereo vision system exploit three camera which is to be calibrated collects environmental information for map building and collision avoidance. The system can combine other sensors, such as sonar and laser in order to better precision and accuracy. In this thesis, the important job is location which determine the performance of follow-up task, such as map building and collision avoidance. In this thesis, we utilize landmark-location to improve the performance of dead-reckoned. We utilize the stereo vision theorem to find the corresponding landmark in left and right image and utilize landmark to calculate the distance between the robot and landmark. We utilize the physics transform to calculate the displacement and rotation between the camera coordinate and the world coordinate, then we can get the location of robot. After locating, we can make the map building. We can get the 3D information of objects from the left and right image which was to be handled to produce 2D map. Then the robot can navigate and record the local map with location information. We can build the environmental map. In this thesis, Our main theme in the design process is trying to develop a low-cost stereo vision system and map building. We add a predictive filter to estimate the location of landmark in next image in landmark-tracking sequence. It makes the algorithm reduce the search area for real time. After locating, we also utilize the physics transform between local map and global map. We can prove the landmark location is effective and accuracy without accumulative error, so it is better to utilize the landmark location than dead-reckoned. Our robot can navigate safely without collision avoidance.