Classification Of Moving Objects For Intelligent Video Surveillance

As an important applition in the video analize of computer vision, especially in the security and transportation surveillance, Intelligent video surveillance plays an important role. Among them, the classification and recognition of moving objects is an important component of intelligent video surveillance system, Through the use of video analysis technology to detect moving objects in the scene for object detection, tracking and classification, Intelligent processing by computer is the basis for the video concentrated and video retrieval.This thesis studies the classification and recognition of moving objects for intelligent video surveillance, processing for the video sequence contains all kinds of moving targets, to achieve classification of moving objects such as pedestrians and vehicles in the scene of fixed camera, the main work is as follows:Using background subtraction detect moving object. Combining the binarized mask with ViBe to establish the initial background, using adaptive threshold method which based on frame level constantly update the background without moving target. And then using background subtraction, binarization, shadow removal algorithms to extract the moving targets, get more realistic picture of prospect moving targets.Extract the global features of the detected moving objects as the feature descriptors. In this paper, it extracts the goals’ global features, combined with the texture such as histogram of oriented gradients, directional derivative features and color space features such as Lab as feature descriptor, and further, we filter characterized image, so that the object can be better adapted to occluded or contained small amounts of shadows among the detection objects, enhance the robustness of the algorithm.It proposes up an improved classification algorithm based on hough forest, using hough forest classifier classify and recognize the moving objects in video. We construct a hough forest classifier based on sample learning theory, using improved hough forest classifier to classify the detection objects.Firstly, training a number of marked samples to generate hough forest classifier, followed by the use of the trained classifier to achieve the classification of unknown object samples, it greatly improved object classification rate.