Distributed Active Embedded Vision System for the Navigation of Differential Mobile Robot

• Main Authors: Chia-Chi Liu, 劉家箕
• Other Authors: Chih-Lyang Hwang
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/99274680177804043835

碩士 === 淡江大學 === 電機工程學系碩士班 === 97 === The navigation for a differential wheeled robot (DWR) in a distributed active-vision pan-tilt-zoom system (DAVPTZS) is developed. The present thesis uses the digital signal processor of TMS320DM642EVM from Texas Instruments Co to be an important platform. The format of color space is YCrCb, where Y denotes the luminance, Cr is the red color, and Cb is the blue color. For the purpose of easy recognition and localization of DWR, the red and rectangular feature is placed on the top of the DWR. In the beginning, the visual information coming from the speed dome is transferred to TMS320DM642 to execute the corresponding image processing, including the segment of Cr component, binary, the removal of noise by median filter, the calculation of the area of image feature, and the computation of coordinate of the center position of image feature. Similarly, the obstacle with blue color and circular shape can be tackled by the above procedure. Recently, distributed control applications within sensor networks become more important. Many of the problems encountered by classic wheeled robots (e.g., localization, high computational power, different software for different kinds of mobile robot, the interference with each sensor) are solved when they are in a distributed network-space. However, almost distributed CCDs are fixed; therefore, the visible region is limited or the number of CCDs should increase to monitor the larger visible area. Although the omni- directional vision system (ODVS) possesses view angle, it contains the following disadvantages. Due to the distortion of image, its image processing is time-consuming and the estimation error (or calibration error) is large. In this situation, a MLP with two-hidden layers are employed to establish the relation between image plane coordinate and world coordinate. The corresponding inputs for the MLP are , , and ; the related outputs are and . After an effective learning, the corresponding MLP is applied to navigate the mobile robot to track a specific trajectory and to avoid the known obstacle.