Research On Abandoned Object Detection In Intelligent Video Surveillance

As more and more attention is being paid to social public security, the role of intelligent video surveillance technology in security precaution is becoming more and more obvious with great potential for development. Abandoned object detection is an important part of intelligent video surveillance which contributes to removing the security risks caused by the unidentified abandoned objects in public areas. Its algorithm design and system implementation are the main research topics of this paper.In this paper, based on the summary of the existing research achievements of abandoned object detection, the key problems to be solved in terms of accuracy, real-time performance and robustness are discussed. By virtue of research on the common methods of object detection, the combination of frame difference and background subtraction is established as the basis of the algorithm implementation.According to the application characteristics of abandoned object detection, the traditional Gaussian mixture background modeling method is improved based on partial updating, associated with an improved three-frame-difference method. The candidate areas of objects are obtained by comparing the foregrounds derived from the two methods. In order to enhance the accuracy of the object area extracted, shadow elimination based on YCbCr color space, morphological processing and connected component analysis are applied, after which the blobs of temporarily static objects are segmented. The pairing of blobs between two frames is achieved by calculating the distance between their centers and the static duration of each object is counted respectively. The disadvantage caused by static humans is removed by applying HOG pedestrian detection intermittently, which maintains both accuracy and processing speed. In order to implement the parallel detection of multiple abandoned objects, a parameter structure characterizing a single abandoned object is designed. The corresponding structure element in the storage container is inserted or deleted in real time with the state transition of an abandoned object, which ensures the independent discrimination and analysis of the characteristics of each abandoned object. In view of the impact on detection due to pedestrian occlusion in the scenes, a method of extracting and matching the FAST local feature of the abandoned object is proposed to maintain the discrimination of its characteristics, which enhances the ability of interference suppression. The algorithm proposed is tested and analyzed using the actual data of surveillance video. The experiment results show high accuracy and anti-interference performance of the abandoned object detection algorithm proposed, and real-time processing can also be satisfied.Finally, an intelligent analytical system of video images used for security surveillance in public areas is designed and implemented in this paper. The principle and function of each module in the system are introduced. The design concept and implementation performance of the software interface are elaborated. The system is developed based on the object-oriented method which is user-friendly with high interactivity, extendibility and practical value.