Research On Human Behavior Recognition In Video Surveillance Based On Neural Network

In recent years, the study of human behavior recognition has become a research focus in computer vision. Identification of human behavior belongs to advanced visual analysis and is an important part in human motion analysis. Human behavior recognition has broad application prospects and potential economic value, mainly in intelligence surveillance, video conferencing, advanced human-computer interaction, medical diagnosis and content-based image storage and retrieval. However, in reality, because of the diversity of life, the non-rigid of behavioral sequence attitude and the fuzzy definition of class, making the study of human behavior recognition to be a very challenging task.As researchers research on the human behavior deeply, making a number of behavior recognition method, according to the difference of behavior description method, it can be divided into motion-based features and shape-based characteristics two types of recognition method. However, the current computer vision technology is difficult to accurately extract characteristics of a person’s movement from the video, and shape features are easily accessed and is not sensitive to the texture, so the shape of features in the behavior recognition is widely used in human behavior recognition. In order to describe the mode of operation of the human body, we use the method of MHI to describe that movement. The motion picture history of human behavior is from image sequences extracted, it is one kind of space-time models. It not only shows behavior area, but also shows how that behavior is occurring, in which each pixel value is the history of sport the function.In this paper, it proposes a human behavior recognition method based on Zernike. By using rotation invariant of Zernike and construcing arbitrary high order moments of the features, Zemike moment recognition is better than other methods, we extract feature by using Zernike moments, image standardization of the history of the movement approach to movement. Boosting algorithm applied to the improved feature classification.s RBF neural network are improved through boosting method, gaining the classifier of a classification algorithm and strong classification performance, it can well improved recognition accuracy.