Study Of Pedestrian And Unusual Activities Detection Methods Based On Video Sequence

Pedestrian detection and unusual activities detection are essential components required by a variety of applications, such as intelligent traffic, intelligent surveillance, image compression, content-based multimedia retrieval and so on. Due to such large number of potential applications, pedestrian and unusual activities can be detected efficiently. However, the pedestrian appearances, numbers, orientations and occlusions, lighting conditions violate and complicated background conditions will deteriorate the performance of detection. In this paper, we present two efficient methods for pedestrian and unusual activities detection.(1) Pedestrian detection method based on histogram of oriented gradient featureIn this part, we realize the pedestrian detection method by using histogram of oriented gradient feature, AdaBoost algorithm, MILBoost algorithm and so on. First we employ HOG feature to describe objects. Then we use AdaBoost algorithm to get seeds of the strong classifiers and MILBoost algorithm to grow the strong classifiers. The last step is cascading all the strong classifiers. Through extensively experiments, the results show that this method is robust and efficient; it has good performance when pedestrians are having different pose and appearance.(2) Unusual activities detection based on Optical Flow method and Spatio-temporal connectionIn this part we employ optical flow method to do unusual activities detection. First the scene is divided into several different regions, for each region we use a monitor to record optical flow features and if there is any unusual activity it will report an alert. For the occurrence of unusual events is continuous, those alarmed monitors must be connected in spatio-temporal space. And we use this method to decide occurrence of unusual activities.At the beginning of each part, this paper will give a brief review of correlated concepts and theories such as Boosting algorithm, Lucas–Kanade optical flow method and so on. Based on these theories, pedestrian detection method and unusual activities detection method are described in details.