This dissertation is supported by the National High Technology Research and Development Program of China under the program “Sport and Entertainment Multi-robot System”. This dissertation is focused on the research of local vision systems to realize relative positioning among humanoid robots. An ARM-based embedded vision positioning system is designed, which can realize the image capturing, image processing and vision positioning for multiple objects. The positioning results are output via serial port. In the designed vision system, S3C2440 produced by SAMSUNG Company is selected as the ARM processor. A CMOS camera C3188A serves as the vision sensor. The software is programmed with C language. The experiments including image capturing, image processing and vision positioning are well conducted for color marked blocks attached on humanoid robots. The measuring rate is about 10 frames per second. The positioning error is less than 30 mm when the target with size of 140210 mm is 1000～2000 mm far away from the vision system. An adaptive threshold setting method based on the average image gray level is proposed. It increases the adaptivity of the vision system to light variation. In order to reduce the computational complexity of the adaptive threshold setting method, grid method is adopted to scan the image and compute the average image gray level. In the different light conditions, the average image gray level and target color value are computed. Then the relation between the average image gray level and the threshold is established with a function. The current threshold can be easily determined according to the current average image gray level and the function. A new feature extraction approach with high speed and precision is developed for color marked blocks. The center of the color marked block is selected as the start point. The 4 groups of points on the block edges are searched in horizontal and vertical directions with specified interval. The line equation is computed for each edge with RANSAC method from each group of edge points. The 4 corner points are determined with the intersections of the lines. Then PnP-based method is employed to estimate the position and orientation of the color marked block in the camera frame. For the designed embedded vision positioning system, many experiments are conducted, such as color block identification and segmentation, feature extraction, objects positioning, object approaching a...
修改评论