Research On Binocular-stereo-vision-based Navigation For An Autonomous Mobile Robot

As mobile robots are widely used in various fields of human life, research onimproving the intelligence level of a mobile robot is highly concerned by the scholars bothat home and abroad. The binocular stereo vision can imitate human eyes, and can get depthinformation of3D scenes by image sensors in a non-contact way. In addition, roboticsrelated technologies, such as large-scale integrated circuit technology, semiconductortechnology and precision machining technology, have developed rapidly in recent years.All those make binocular-stereo-vision-based mobile robot navigation become the researchfocus of intelligent robots field. Therefore, study on key technologies of the visualnavigation has theoretical and practical significance.This thesis focuses on some key technologies of binocular visual navigation, whichinclude stereo calibration, image acquisition and image preprocessing. And some researchon the rapid and accurate localization of spherical objects in indoor environment is alsodone.Firstly, a binocular-vision-based omnidirectional mobile robot is designed to verifyvalidity of the theoretical research. Hardware architecture of the mobile robot is provied,and main components of the robot are introduced one by one.Secondly, camera imaging models are systematically introduced, and from whichparameters for the stereo camera calibration can be derived. Then, the main principle ofZhang camera calibration method is outlined. And the purpose of the stereo calibration,which is to calculate the rotation matrix and the translation matrix containing the geometricrelationship between two cameras, is analyzed. Finally, open-source computer visionlibrary named OpenCV is employed to calibrate the binocular stereo camera in VisualStudio2010IDE, and experimental results indicate that the reliability of the calibrationalgorithm.Subsequently, a DirectShow-based image acquisition system used in the mobile robotis presented. Gauss filtering algorithm is used for smooth filter, and Canny algorithm foredges detection in image preprocessing. An improved symmetry-based Hough transformalgorithm for circle detection is proposed. Experiments are carried out to detect the objectwith circular features in indoor environment, and their results are prospective. Finally, the spherical target location method based on the principle of parallax is given.The mobile robot location and tracking program is written by using OpenCV andDirectShow in Visual Studio2010IDE based on MFC. The experimental results verify thework. And the factors which affect accuracy of location and tracking are also analyzed.