In Robot Vision System Based On Neural Network Image Recognition Technology

With the development of computer and communication technology, each module, such as environment sensing, direction deciding, route planning and movement controlling module in the system of intelligent robot, is resolved respectively. So these modules can work simultaneously and the whole system' s real-time and stability are improved.As for environment sensing, vision processing is key factor. On the base of researching the theory of machine vision and as per specific recognition task assigned to the robot, the paper discusses image processing method by using digital image processing technology, and then the recognition of processed images by using neural networks. Intelligent robot can recognize different objects or image commands as per request. In this system, robot will identify two kinds of image commands: character and gesture . In order to implement this task, the paper studies two different processing methods, with relevant software programmed with VC++6. 0.Details are as follows:Firstly, basic principle, current development, application of machine vision and image processing & recognition are described in details, as well as necessity and feasibility of their integration.Secondly, several methods of image recognition, including Statistical Pattern Recognition, Syntax Pattern Recognition, Neural Network are studied in this paper. Neural network is used widely in many fields as it' s strong self-adapting, self-learning and error tolerance abilities, with which image recognition is implemented. The paper comprehensively deep researches on some knowledge about artificial neural network concerning its principle, structure design. Problems and correcting methods in application are also proposed.Thirdly, different processing methods for different images are described in details. For character images, processing method mainly includes gray scale transformation, two value transformation, pick-up, emendation and zoom of character. For character image tilt, edge search is used. According to search result, image is verified by using Hough transform. For gesture images in YUV field, firstly we divide hand image from complex background, then make further process, then make recognition by using neural network. We obtain good recognition result in this paper for both of these two images.Finally, some pictures obtained from experiments are provided and experiment analysis are given. The shortage of the paper, the direction of further amelioration are also indicated.