Streaming Video Key Frame Extraction And Identification String Recognition

More and more attention is being given to the computer visual technology which is mainly used in ITS(Intelligent Transportation Systems). This technology has three important techniques to be resolved: the detection of the key frame in video stream, character segmentation and extraction, recognition of character symbol. This text has done some research on these three fields and supplied the designs of the front two fields related to their special application. Character recognition is the emphasis discussion of this text.The method put forward in this text is the combination of contiguous frames difference and background difference.Video stream is divided into four periodic states. We could detect the key frame precisely by judging the current, the former and later the states.Using the segmentation in OCR system for reference, character segmentation in this research finds the solution to the problems of character declining and character touching which affect the segmentation quality.Character recognition belongs to classic pattern recognition. Traditional template matching recognition could be realized easily but its recognition ratio and anti-jamming capability are low. This research could make up that gap to some extent by using NN technique. Firstly extract the character feature by the use of statistic pattern recognition theory and then create large sums of aberrant character samples by PC simulating, research the statistic quality of character feature in standard space and aberrant space. On the basis of that, create different linear recognition machines for characters in different spaces imitating the perceptive machine model in NN theory. The linear machines we designed have got better effect in practical application.