Research On Some Computer Vision Techniques Of Surveillance System

Nowadays, video surveillance system has been widely used in various public places, the monitoring system itself is in the same period of rapid growth, showing a larger-scale monitoring system, using cameras and hardware systems become more complex and also increasingly monitor area greater. At the same time monitoring system for content analysis and processing requirements are also increasing, moving from simple monitoring to intelligent image processing and analysis of the contents of the direction.Image content analysis is to study the image content, including but not limited to use of a variety of image processing technology, it is more inclined to image content analysis, interpretation, and recognition. Image analysis combined with the general use of mathematical models and image processing technology to analyze the underlying characteristics and the upper structure, to extract certain intelligence information.This paper introduces:1) Using on-line boosting algorithm for structure detection of target objects. Different from the classic AdaBoost algorithm, we have introduced an concept of selector. Training a selector means that its associated weak classifiers are trained (update), and also pick out a best weak classifier. Through experiments, we have the feasibility of detection algorithms, and to explore to improve the performance indicators.2) License Plate Recognition for Intelligent Transportation System is an important topic to many researchers and manufacturers. Through a large number of theoretical and experimental analyses, a set of relatively completely license plate location algorithms are established. The algorithm we propose can adapt to indoor and outdoor complex lighting conditions. In the more accurate and rapid localization algorithm, license plate recognition we turn on the two other processes:character segmentation and character recognition are explored and discussed in trade-off between the recognition accuracy and speed.3) As the network applications and the rapid development of multimedia technology, making the image data showing explosive growth. Content-based image retrieval technology because of its important practical research significance and great value get more and more attention. We present the most popular types of image descriptors to find the most suitable method for the expression of "similar" way. Among them, we focus on the SIFT features and VLAD aggregation characteristics. Experimental data on the performance of these image descriptors do a more comprehensive analysis. In order to overcome the enormous set of large-scale image calculated pressure, we studied and successfully implemented the structure of ANN based on the KD Tree Algorithm.