Study on Spatial Localization of Pill Box Gripping Point Based on Binocular Camera

Abstract

In order to realize the accurate grasping and localization of the target object by the robotic arm, this paper proposes a three-dimensional spatial coordinate solving method based on binocular vision. By establishing a binocular stereo imaging model, the geometric mapping relationship between spatial points and pixel coordinates is deduced, and a binocular calibration experiment is designed to obtain the camera internal reference and distortion parameters. Aiming at the assembly error of industrial cameras, a stereo correction algorithm is adopted to realize the image polar line constraints, combined with the RGB-D depth image alignment technique to eliminate optical aberrations, and construct a sub-pixel level aligned visual perception system. On this basis, the deep learning pill box detection algorithm is fused to realize the 3D coordinate solution of the grasping target through feature matching. Experiments show that the method can control the localization error within the permissible range of robotic arm grasping.