Research On AGV Localization And Mapping Based On Binocular Camera

With the rapid development of automation level and sensor technology,people's dependence on intelligent products was increased.AGV are widely used in various occasions,such as unmanned driving,logistics,AR/VR and so on.Navigation is the core of the AGV system.Effectively solving the positioning and navigation problem is considered to be one of the key technologies for mobile robots to realize artificial intelligence.The real-time positioning and map construction based on vision was an important research in the field of robot navigation in recent years.Aiming at the visual navigation requirements of mobile robot in an unknown environment,a visual odometry based on binocular camera was proposed.Firstly,for the problem of feature redundancy,the multi-threshold FAST image segmentation was introduced to improve the ORB algorithm.The FLANN and RANSAC algorithm was used to reduce the mismatch.Then,the stereo matching algorithm based on descriptors instead of gray was proposed to obtain the depth.To get the precise position,the least squares that provided initial values was raised,and the camera motion was estimated using the matched feature position.In the map reconstruction,the Bundle Adjustment based on back-end optimization was used to optimizing the feature points and camera poses with the sparseness and marginalization of the matrix,in order to reduce the reprojection error during the motion.Experimental results demonstrate that the proposed visual odometry based on binocular camera can improve real-time performance and location accuracy,which can effectively restore the environmental information around the map.