Real-time Detection and Classification of Machine Parts with Embedded System for Industrial Robot Grasping

In this paper, a real-time machine vision system is designed for an industrial robot to grasp from an assembly line a class of machine parts which are similar in the general shape but different in details. In order to get real-time performance, the system is implemented on an embedded image card with an FPGA (Field Programming Gate Array) accelerating the computation. The method can be divided into two stages. First, the holes and edges of the machine parts are detected from each frame with the FPGA. Then a DSP (Digital Signal Processor) chip on the image card performs the rest of the computation by identifying the location and type of each of the machine parts in the image based on the information of all the holes and edges. A rotationally adaptive edge-based template matching technique is used in our method, which not only reduces the amount of computation but also provides robustness against illumination changes. Experiments demonstate that the machine parts can be located accurately under arbitrary in-plane rotations and can be classified correctly according to the details in their shapes. Our system can run with an industrial camera at a resolution of 640x480 and a speed of 50 fps (frames per second) or higher.