Research Of Real-time Character Recognition Technology In Video Surveillance Based On Android Platform

Android is one of the most popular smart-phone operating systems, it is an open source mobile phone platform published by Google based on Linux platform. With the3G era, the development of mobile application terminals has become focus of modern people’s attention. The smart-phone operating systems come into being, which makes the related applications more and more important. The current Android intelligent mobile phones and pads are equipped with sophisticated and high-profile cameras, which can capture text information in people’s life whenever and wherever possible. Based on Android platform, the research and development of character recognition software become more and more necessary in accordance with users’ growing demands for character recognition.According to the application requirements of real-time character recognition of single line printing English letters and numbers in simple backgrounds on Android platform, this paper uses Android developing technologies and a variety of image processing methods to research on video processing, character location, tilt adjust, character segmentation and character recognition. As a result, the real-time character recognition system based on Android is designed and realized. The main research works and results achieved are as follows:(1) Research on pretreatment of real-time Android video. The main work includes a variety of pretreatment processing methods and related algorithms, such as converting YUV420data to RGB format, gray, binaryzation and edge detection. Above all, Android NDK framework is used to improve processing efficiency. Experiments show that, using this method just cost20.58%of the time cost by using general methods under the premise of extracting character region.(2) Research on the method of real-time character location based on ROI motion detection. In consideration of the features of continuity and similarity of video frames, this paper uses this method to track device position, so the repeated locating process of the same character is omitted. As a result, the locating efficiency is improved. Experiments show that, using this algorithm character can reach a locating accuracy of95.67%. Comparing with the ordinary method, the average processing rate is improved by40%.(3) Research on the method of character segmentation based on two-dimensional image projection. In order to improve the accuracy of character segmentation, this paper uses a method based on two-dimensional projection. Through the analysis of horizontal and vertical projection segmentation of the character images, the key points are found out so that the single character is divided for template matching.(4) Research on the method of character recognition based on combination of multi-temple matching method and error threshold selected method. With the research and analysis of template matching method and error threshold method, they’re combined to ensure both the accuracy and the recognition rate. Experiments show that using the proposed method in this paper reaches a recognition accuracy of94%and the average processing rate is improved by58.4%compared to the general method.