Resarch On Pedestrian Detection And Tracking Technologe Based On The Video

It is the rapid development of computer technology, computing capability has been greatly improved, making use of computer realization of human visual function in the computer field become one of the hottest topics. As a hot research topic and a foreword direction in field of computer vision, Pedestrian detection and tracking technology has important application value and practical significance in many aspects of intelligent traffic, demographics, public safety and intelligent monitoring, However, because factors of pedestrian posture differences and complexity of the environment, Pedestrian detection and tracking technology is still facing many difficulties, it needs further study to provide the necessary technical support for various fields. Considering these difficulties, aim at improving pedestrian detection rate and pedestrian tracking stability, the main content of this research can be summarized as follows.1. This article summarizes the image features used in the field of pedestrian detection and tracking image features, such as HOG, Haar, Edgelet and color characteristics, and feature extraction methods are described in detail, as well as analysis their strengths and weaknesses.2. This article extracts region of interest based on Gaussian mixture model and based on frame difference, The foreground region is firstly extracted through these two methods. Then the region of interest is obtained by processing the regional connectivity. And this article also study the pedestrian detection algorithm based on a single feature classification.3. Considering the region of interest and multiple features, we improve the traditional pedestrian detection algorithm based on Haar features AdaBoost classification. it uses two classifiers to detect pedestrians within the region of interest, firstly the algorithm roughly detect samples by Haar feature classifiers and then detects candidate pedestrian target by Edgelet feature classifier.Finally the article also compares the performance of the proposed algorithm and the traditional algorithm.4. In order to improve pedestrian tracking stability, the verification step was added in the traditional particle filter pedestrian tracking based on color features,firstly, the pedestrian is detected using pedestrian detection algorithm to get the initial target template, Then the target is tracked based on particle filter utilizing color histogram, Thereby expanding the scope of forecasting results, the article use pedestrian detection algorithm in Chapter III of this article to validate the predicted target, and obtain final pedestrian tracking results.Finally, we simulate and compare the improved particle filter algorithm and the traditional particle filter algorithm.