Research On Detection And Tracking Technology Of Moving Object In Intelligent Video Surveillance

Intelligent video surveillance is an important branch of computer vision. It is not only consistent with the development trend of information industry, but also represents the future direction of surveillance industry, which becomes a hot research topic in recent years. Moving object detection and tracking are two core technologies in intelligent video surveillance and the foundation of following-up advanced processing and application. This thesis mainly focuses on moving object detection and tracking, proposing appropriate solutions in view of the existing problems under the analysis of the existing research results.The main contributions of this thesis are as follows:In moving object detection, an introduction of three common algorithms is made as well as the advantages and disadvantages of frame difference and symmetrical frame difference are analyzed. An improved moving object detection algorithm based on frame difference is proposed. The test proves that the improved algorithm increases the accuracy and integrity of moving object, getting more attributes of moving object, which not only achieves object position but also obtains object moving direction.In moving object tracking, for the shortage of traditional Mean Shift tracking algorithm, we get an improved object tracking algorithm. It establishes initial object motion template based on features which is characterized by the weighted gradient direction histogram; for positioning to the object, the algorithm firstly predict the possible object position in the next frame using Kalman filter, then apply the Mean Shift algorithm to search the object position using the predicted position as the initial estimate; in order to track object, we propose a template updating method based on moment feature which realizes moving object template adaptive updating. Experiments show that this algorithm overcomes the shortage of Mean Shift which is vulnerable to the background similar to object color. Moreover, when the external environment changes, the improved algorithm can still accurately track the object because of the timely update of the template.Many tracking algorithms can only track the general area where the object locates. This thesis incorporates contour detection method into the improved tracking algorithm in order to implement the precise tracking based on moving object contour and favour the continuing study such as object identification and classifying. Combining with the projection method and region growing algorithm, the improved algorithm realizes the initial goal of automatic and semi-automatic acquisition based on moving object detection algorithm. The deficiency caused by the need of achieving initial goal artificially of traditional Mean Shift algorithm is overcome and the intelligence of tracking algorithm is improved.