The Research On Pose Estimation Of Automobile Steering Knuckle Based On Depth Camera

In recent years,with the vigorous development of robotics and machine vision technology,the traditional manufacturing industry is undergoing technological upgrading.More and more robots are entering the factory floor,taking on various tasks.Using vision sensors to cooperate with robots to complete various tasks has become an important part of the field of industrial automation.The traditional vision-guided robot positioning is mainly oriented to 2D and neatly placed scenes,and it is difficult to solve the positioning and grasping of workpieces with different postures.This article combines the actual loading and unloading in the process of automobile steering knuckle processing,using 2D image processing and point cloud processing to realize the positioning of the steering knuckle.First,it analyzes the machine vision technology and different methods of 6D pose estimation based on point cloud.The overall plan of steering knuckle 6D inspection is constructed,the imaging principle of depth camera is studied,the camera’s distortion is calibrated,and the point cloud is generated according to the calibration result.Secondly,in view of the deviation in the size of the center hole of the steering knuckle,it is impossible to directly use the point cloud registration technology to achieve indirect positioning of 3D coordinates.The design algorithm uses 2D image processing to determine the pixel coordinates and the depth camera index depth value to determine the3D coordinates of the grab point.Thirdly,in view of the different attitudes of the steering knuckles and the need to estimate the attitude when grasping,it is proposed to use the SAC-IA algorithm to initially register the steering node cloud to be detected with the template steering node cloud,obtain the initial conversion matrix.The NDT algorithm performs optimized registration,and the final matching score is 27mm~2,which determines the precise posture of the steering knuckle to be grasped.Finally,carry out software development and system testing.On the Windows system platform,the VS2015 compiler,Open CV vision library and MFC are used to design the steering knuckle pose detection system software,and the Intel RealSense D435i camera is used to collect the knuckle pictures for visual processing and analysis in the software.