关键词: 零件识别, 装配监测, 像素分类, 深度图像, 随机森林分类器

Abstract: Aiming at the problem of parts and assemblies identification and monitoring in mechanical product assembly,a method of parts recognition and assembly monitoring based on depth image and pixel classification was proposed.The assembly depth image labeling sample set was constructed,including the synthetic labeling sample set and the real labeling sample set.The depth difference feature was extracted and the pixel classification of the depth image was realized by using the random forest classifier to obtain the pixel prediction image.By comparing the pixel prediction image with color label image,the parts of the assembly could be identified;by comparing the pixel prediction image of state to be measured with the correct assembly pixel prediction image,the assembly process could be monitored.To improve the classification accuracy of pixel points on the edge,the edge factor was introduced.The experimental results showed that the method had high accuracy,real-time performance and robustness,and had certain application value in the field of assembly maintenance guidance,assembly monitoring and automatic assembly.

Key words: part recognition, assembly monitoring, pixel classification, depth image, random forest classifier