Detection and Location of Sheet Metal Parts for Industrial Robots

Abstract

This paper presents a multi-object recognition and location approach based on a 2D vision for Sheet Metal Parts. This novel proposed approach allows to identify several texture-less parts to be manipulated using a KUKA KR6 R900 sixx robot arm. The particularity of the suggested method is to build up a process able to recognize this kind of parts characterized with insufficient details to be trained with. The proposed solution overcomes detection problems related to parts appearance variability due to changes in color and contrast under different lighting situations. PatMax tool was used for workpieces recognition and to determine their location. PatMax and PatQuick algorithms were tested with a set of runtime images of 144 different samples. All the parts have been successfully recognized then sorted. The experimental results confirmed the performance of Pat Max and the minimum recorded score was 95%. Fit error scores with PatMax were close to 0 while coverage scores were close to 100%, indicating a good model-pattern fit. The clutter score was calculated based on the proportion of the extraneous features present in the found object compared to those of the trained pattern. This is an assessment of the degree of features absent at the execution level. Based on the obtained results, 85% of detections had a zero clutter score.