Research On Binocular Vision Localization Of Indoor Mobile Robot Based On ARM

Binocular vision localization of mobile robots based on embedded system in indoor environment is a research topic that attracts much attention at present.Traditional embedded platform can not support high data density and high real-time visual positioning services,which can not solve the contradiction between computational complexity and positioning accuracy.To solve this problem,ARM Cortex-M3 is used as the slave machine to control the motion of the robot.The Jetson TK1 development board with high performance-price ratio is selected as the host computer to process the optimized visual localization algorithm,and self localization and target location of the robot binocular vision system is realized in indoor environment.The main research contents and achievements of this paper include four aspects:(1)In view of the limited indoor environment,the omni directional mobile platform is built as the chassis of the robot.By establishing the inverse kinematics model of the four-wheel omnidirectional mobile platform and designing the chassis hardware circuit,the precise closedloop control of the mobile platform is realized.(2)By analyzing the principle of binocular stereo vision measurement,a mathematical model of binocular stereo vision is established.Aiming at the problem of camera imaging distortion,distortion correction and level line are used to correct the image by calibration parameters that are obtained through calibration experiments.The stereo matching algorithm based on feature points is analyzed and compared.A parallel acceleration scheme is designed in order to solve the real-time problem of stereo matching algorithm on embedded CUDA platform and the effectiveness of the scheme is verified.At the same time,an improved method for matching accuracy of SURF algorithm is proposed,which has achieved good results.(3)According to the framework of visual SLAM,a robot vision positioning scheme is designed.In view of the shortcomings of ORB-SLAM2 open source scheme,optimization is made at the code level from the perspective of embedded engineering application.Visual interface and redundant code are deleted,dictionary loading speed is increased,and application interface based on ROS system is developed.At the algorithm level,the point cloud data is converted into two-dimensional trajectory map to facilitate the analysis of the actual trajectory of the robot.(4)According to the characteristics of embedded platform,a three-dimensional target location algorithm is proposed by combining color space,morphology and stereo matching in image processing.The experiments of manipulator positioning and grabbing show that the algorithm has better positioning accuracy and real-time performance for cubic objects with significant color features.Finally,the overall verification and data analysis of the robot vision localization algorithm prove the effectiveness of the system design in the indoor environment by AGV experiments.