Visual Detection System of Automotive Parts Attitude Based on Deep Learning

In the automobile assembly process, pose detection technology is the key technology in the assembly system, which plays an important role in the assembly work and will directly affect the assembly results. In order to improve the detection intelligence and accuracy, the deep learning vision technology is applied to the attitude measurement of automotive parts assembly, and a method of extracting key points by deep learning is proposed. Different from the traditional methods, the deep convolutional network has strong robustness and versatility. Therefore, this paper proposes a joint detection method for double convolutional networks. First, the deep convolutional network is used to predict the key points of the part. Then, the recursive convolutional neural network is used to optimize the prediction result. Finally, the key point coordinates are substituted into the PnP algorithm to check whether the part pose conforms to the standard. The simulation results show that the method can accurately realize the target attitude detection.