Research On Technology For Auto-recognizing Welding Joints Based On Neural Networks And Computer Vision

Auto-recognizing the welding joint based on computer vision is an important aspect in the research field of the welding robot. The conventional teach mode for the welding robot is of many disadvantages such as the great workload and low efficiency. After we have explored some existed methods, we put forward a new approach of auto-recognizing the welding joint in this paper, which bases on computer vision and neural networks.In this paper we introduce the principle and implement scheme from our method firstly. Then, some correction methods for the distorted image are presented, in which the distortion model needs to be set up and have a complex solving process including complicated calculation and numeral error. So we give a new correction method based on neural networks in this paper, which has been better applied to distort the die carrier images. Many good experimental results from our method have also been obtained. And then, we preprocess the die carrier images with the techniques of intensifying contrast, image segmentation and edge detection based on neural networks. We have also improved the Hough transforms algorithm for the two ends of the line cannot be detected and the lines are detected repeatedly with a low precision by using the conventional Hough transforms. Finally, the improved Hough transforms algorithm is utilized to extract lines from the die carrier images.On the base of the theory analysis and research, the whole mechanical system and electrical system for welding robot have been established and the control software is developed with VC6.0.