Research Of Outdoor Robot Monocular Visual Odometer

In the field of robot, image is widely researched and used, because it can accurately describe the scenarios, abundant environmental information. Image information is also called the important access point of environment perception and identification. Image infonnation is attached great importance in Robotics and various fields because it is indispensible in the visual system of mobile robot navigation technology. Robot navigation, first of all, the first problem is the real-time positioning problem, accurate positioning is essential to the robot navigation. Because of its low cost, fast processing speed, simple operation and high accuracy, monocular visual odometer has become an important option in visual navigation.This paper first summarizes the mobile robot navigation technology, the research and the use of field and the composition of the visual system, and describes briefly the present situation of the domestic and foreign, and then visual odometer is introduced combining the theory of robot visual and visual navigation technology.Because of its characteristics such as non-contact, intelligent, fast detection, using visual method to detect robot position is widely studied in recent years. Based on Poineer3platform, we have carried on the outdoor video image acquisition. With the method of local fractal dimension and neutral network, we achieve the recognition of the ground and identify the outdoor traffic signs. Regard the robot speed as the ground movement speed, Horn-Schunck optical flow method is used to calculate optical flow of the ground. And then by the relation between the image coordinate system and the robot coordinate system, the distance is measured between the research object and the robot. According to the relationship of velocity of movement and optical flow velocity, the optical flow velocity is converted to the speed of robot. This method is feasible and easy to operate with a fast inspection speed.Finally, detecting and matching two adjacent frames collected from outdoor environment by using the SIFT algorithm, and then eliminating the false matching points with the method of the RANSAC algorithm. Achieving the robot positioning according to the visual model of the odometer.The experimental results show that the positioning precision using this method is higher than the traditional odometer, and it is cheaper than binocular and laser. It provides a solid foundation for the further study and work to fuse the optical flow method and the feature point matching algorithm of monocular visual odometer.