Intelligent Video Surveillance And Retrieval System Development

With the development of information technology, intelligent video surveillance and retrieval technology had been widely researched. Compared with traditional video surveillance, intelligent video surveillance and retrieval system involved a number of interdisciplinary researches in the field of machine vision, pattern recognition, and artificial intelligence. This system made use of intelligent algorithm to assist people to complete the monitoring and retrieval tasks. In this paper, the work completed as follows:1. Intelligent video surveillance storage systems; 2. Intelligent surveillance video retrieval system. This part was divided into three subsystems of object detection, object tracking and object retrieval. Finally, an intelligent video surveillance and retrieval system was designed by this paper.For traditional video surveillance system, there was the problem of redundant information in continuous video recording and storage processing. A lot of storage space was consumed and information retrieval efficiency was reduced for this traditional system. In this paper, intelligent video surveillance storage method was designed to solve this problem. The video information was recorded and saved, when moving objects was detected by the system. This method could save storage space and facilitate the subsequent search and retrieval.The intelligent retrieval system of video surveillance was designed by this paper, it could establish retrieval database by detecting moving objects and search the information for user’s needs. The main job as follows:1. Improved Vibe object detection algorithm was proposed. Firstly, compare the histogram similarity of foreground, ghosting and each neighborhood to inhibit the ghost. Secondly, use the characteristic of shadows brightness lower than the background’s to remove the shadows. Finally, use the morphology algorithm to fill the blank area. Compared whit the Gaussian mixture model and traditional Vibe algorithms, the experiment showed that this improved Vibe algorithm Fl-measure was respectively increased by 24% and 5%, and had good real-time performance.2. Multi-object visual tracking algorithm based on fragments multi-feature adaptive fusion was proposed. This algorithm adaptively fused color, texture and edge features, and combined with Kalman filter to predict objects in the case of occlusion. The experiment showed that the rate of occluded object recognition was 95.3%, the average processing time for each frame was 36.2ms.3. The retrieval system based on content and semantic was proposed. Combine with color histogram and SIFT feature, the average recall rate was 86%, the average precision rate was 88%, and the average retrieval time was 13s. The semantic retrieval recognition rate was 87%, using SVM for image classification and optimizating parameters.Finally, this paper designed and realized the intelligent video surveillance and retrieval system. On the intelligent video surveillance storage, the savings rate was 30%. On object detection and tracking, the algorithms met real-time requirements. On object retrieval precision rate could reach more than 86%.This system is able to achieve satisfied results.