面向全自主智能机器人的双目视觉技术研究-计算机科学与技术专业毕业论文.docxVIP

哈尔滨工业大学工学硕士学位论文 哈尔滨工业大学工学硕士学位论文 -II- -II- ABSTRACT The information of fully autonomous intelligent robots is processed by the chip-on-board completely, so most of this are designed by embedded system. There is a wide gap on processing speed and storage capacity for fully autonomous intelligent robots which is designed on embedded system relative to ordinary PC, so let the image processing algorithms run better in embedded systems is a big challenge. Currently, almost all of the fully automous intelligent robots apperceive the surrounding environment by using a single camera. Though the monocular vision does have its advantages relative to the binocular stereo vision, and it exists significant deficiencies also. This thesis contrast the monocular and binocular visual performance from the region of interest segmentation, distance accuracy, time complexity firstly and summary the following conclusion. Monocular vision has small amount of information and only deals with one image each time, so it has fast arithmetic speed and often be used in the simple scene. But the 3D information of the actual scene is lost in the process of projection. At the same time the monocular vision system stringent requirements on surrounding environment, such as light, color, saturation. Some unendurable judgement mistakes will appear if the scene becomes a bit more complicated. Relatively, binocular stereo vision overcomes these problems very well, meanwhile it increases a lot of computation, so the real time autonomous control decision of fully autonomous intelligent robot which is designed on embedded sytem becomes a challenge. In this thesis, whith demand of the need to determine the coordinates of the region of interest for intelligent robot, proposed a efficiency coordinate calculation method which based on judge the longest straight-line, this method is robustly and hight accuracy. By in-depth analysis and research the stereo matching through a large number of experiments, decided