Study On Identification Of The Traffic Of Freshwater Based On Computer Vision

This paper is the focus of the project - "Study on the application of freshwater sea-route multi-vision sensors information fusion technology". The research on identity recognition of the traffic of freshwater based on computer vision is that the captured license plate images of the traffic by vision sensor can be automatically recognized using computer vision techniques, and some basic information of the traffic are obtained. At present, a law of management of the license plate of minitype ship has been developed by the government; it requires all the traffic of freshwater must have the license plate before 1/7/2007, which provides a basic frame of reference for identification of the traffic. This Paper is mainly about three parts: license plate automatic location and segmentation, image preprocessing and character segmentation, character recognition.(1) Research on license plate location and segmentation. According to the characteristics of license plate image, a color segmentation method based on the CIEL~*u~*v~* uniform color space is presented, by transforming the RGB space into that of CIEL~*u~*v~*, then clustering color by Euclidean distance and eliminating noise by mathematical morphology, finally the plate region is extracted.(2) Research on image preprocessing. A series of preprocessing techniques are used to improve the image such as gay image, gray stretch transformation, histogram equalization, spatial smoothing filter, iterative threshold binarization, and slant images are corrected by a method based on Hough transform and K-means-based line fitting algorithm.(3) Research on character segmentation. A novel method of character segmentation based on combination of template matching and projection is proposed. It can be efficient and robust in dealing with uniform-illuminated, constrast-poor, declining, dirt, character-touching and character-broken number plates.(4) Research on character recognition. Template matching and BP neural network are used to distinguish the characters. We use a weighted voting rules to integrated the output which recognized by three template matching classifiers, then the recognized rejection characters based on BP neural network. The result shows that it is more effectively in character recognition.The related research in this paper have been designed and implemented in the intelligent surveillance system for freshwater sea-route based on multi-vision sensors information fusion technology, Experimental result shows that this method we proposed is accurate in ship license plate location, segmentation and recognition, having a good performance.