Pedestrian Detection Technology In Moving Background

The intelligent video surveillance system is a new demand for the security and intelligence of urban residents in the area of family housing, residential community, parking lot and so on, and moving target detection and pedestrian recognition is the core part of the intelligent video surveillance system. How to detect and recognition of the human target in the moving background caused by the camera movement is still faced with many challenges. In this paper, multiple moving target detection and moving human target recognition in the moving background is studied. The main research content is as follows:After studying the detection algorithms on multiple moving target in the moving background, a multiple moving target detection algorithm based on sparse optical flow and improved DBSCAN clustering is proposed. In the algorithm, Harris-Sift is used to extract and match image feature points, then the motion vectors of the feature points of the adjacent frames are transformed into the corresponding optical flow chart, the improved DBSCAN clustering algorithm is used to complete the clustering on the optical flow graph, Thus, the moving object is separated from the moving background accurately, and the estimated area of each moving target is obtained.After studying the detection algorithms on the human motion target in the moving background, The algorithm of moving human recognition based on the human geometric features and skin features is proposed. When the moving target meet the geometric features of the pedestrian, but not meet the pedestrian face skin color model, using HOG (histogram of oriented gradient) feature with SVM (support vector machine) classifier for pedestrian detection method, the target will be identified as the human body by the estimated regional input SVM pedestrian detection classifier.The implement of the algorithm of moving human detection and recognition in the intelligent video surveillance system was introduced, and then the development of dynamic libraries of the moving human detection and recognition algorithm under the Android embedded system was described in detail. Finally, this paper applied the dynamic library to the intelligent video surveillance system.