Research On Moving Object Detection And Recognition In Video Sequences

Moving object detection and recognition in video sequences is an important part of intelligent video surveillance system, and it is also a research hotspot in the area of image processing, computer vision, pattern recognition. It has an important research value because of its wide applications in military guidance, security monitoring, intelligent transport, etc.This paper mainly aims person and vehicles to investigate the technology of moving object detection and recognition under stationary background. Moving object detection, feature extraction and object classification is mainly researched in this paper. The main research work can be summarized as follows:In the aspect of moving object detection, the principles and advantages and disadvantages of the current several common detection methods are discussed and analyzed. Some kinds of algorithm such as background subtraction method based on three frame differencing, background subtraction method based on RGB color space and background subtraction method based on Gaussian mixture model with neighborhood information combined are achieved, also, speed, accuracy and adaptability of these algorithm is evaluated. In view of the existent interference of shadow in the process of moving object detection, a shadow detection algorithm by integrating two features, which combines the brightness ratio characteristic of HSV color space and the logarithm ratio invariability of neighborhood shadow pixel to detect shadow, is proposed on the basis of analyzing the characteristics of the shadow, this method with good robustness can not only automatically set key parameters, but also adapt to various environments. To remove the existent noise region after moving object detection, some post-processing algorithm such as filtering, morphology processing, connected components analysis are researched and achieved, and the result is good.In the aspect of feature extraction, Hu moment feature and NMI (Normalized Moment of Inertia) feature is studied in this paper, and their performance is analyzed from the sides of invariance and separability. A geometrical feature base on wheel positioning by Hough transform is designed in view of vehicle object, and this feature, which can be applied in identifying vehicle's model accurately, will be used widely.In the aspect of object classification, related theory of support vector machine is explored and researched. A classifier based on support vector machine is designed, and quantum genetic algorithm is applied in the optimization of the parameters of support vector machine in view of determining these parameters by trial and error in the past, and the classifier's performance is improved.At last, according to the particular scene of daytime outdoor some roads, a moving object detection and recognition system is designed and constructed based on our algorithm, which uses OpenCV technology based on VC++6.0 development platform, the system is demonstrated and analyzed by system experiments, and the real-time performance and effectiveness of the system is verified.