In order to solve the problem that traditional industrial robots cannot sort and grasp multiple workpieces without positioning positions, a machine visionbased parts sorting and loading and unloading system was built.The system could be used to sort aluminum and plexiglass parts more accurately, recognize the pose of the workpiece, and finally convert the result to the robot’s world coordinate system to accurately grasp and load the workpiece. Among them, for the sorting of workpieces, a method of identifying and sorting through bar codes was proposed, which was simple, effective, fast and convenient. For the pose recognition of transparent plexiglass parts with low contrast to background, an adaptive parameter adjustment linear transformation method was proposed to accurately identify the pose information of the transparent parts. The experiment proves that the system has good accuracy and robustness.