| Intelligent video surveillance system uses the computer vision technology to analyze thecollection video information from camera, filters out irrelevant information, and extractsuseful information to monitor persons to process, thus which has warning, prevention andactive monitoring function. In the field of intelligent video surveillance, much attention hasbeen placed on tracking people, in security, advanced man-machine interaction, crimedetection and medicial diagnosis, etc, which has broad application prospects and potentialeconomic value.This paper studies how to accurately detect and track the targets of surveillance video inreal-time. Based on existing results, the relevant technology of moving target detection andtracking has been improved accordingly.Firstly, the video is susceptible to noise in capture process, before moving targetdetection, which needs to be pretreated. This paper uses an improved vector median filteringalgorithm to remove the noise in the color video image, and gives the experimentalsimulation.Secondly, several common methods of moving target detection are compared andanalyzed. In view of those methods are easily affected by noise, light and prone to false goal,so a new video object segmentation algorithm combining the edge detection and inter-framedifference is proposed in this paper. And, the motion detection results are easy to appear somebad phenomenons, such as shadows,"hole", and edge discontinuous, and so on. Therefore,further research is needed on the shadow removal, edge connection and morphologicalpost-processing.Finally, for the specific application, this paper selects a feature-based tracking algorithm.More attention is focused on feature selection, description, extraction and template updating.At the same time, in order to improve the performance of the algorithm further, Kalman filteris introduced before the feature tracking. Through forecasting to locate the moving target,then identify and track the moving objects within the location area. |
PDF Full Text Request