| The world of big data has arrived,the traditional monitoring system has been difficult to meet the future needs of society.Intelligent video surveillance system as a product of the Data Technology Age which applied in various fields.Tracking as an important part of the intelligent video surveillance system,which plays an important role in the subsequent work,such as behavior recognition,anomaly behavior analysis,etc.However,due to the complex of the reality environment,finding a robust tracking algorithm is one of the problem to be solved.Therefore,combining with the knowledge of computer vision,this paper has a deep research on visual tracking and its application.We propose corresponding improvements and solutions based on the existing algorithms of visual tracking.Meanwhile,we propose a loitering detection algorithm which based on tracking trajectory.Specific research contents are described as follows:Firstly,the appearance model may show significant changes which causes by the deformation or the change of camera angle during tracking.This paper investigates the contribution of saliency map for object tracking,and proposes a saliency detection method which is combined with the location information.We redefine the context prior model by saliency map instead of image intensity which follows the focus of attention model better.In addition,we propose a tracking algorithm which combines online spatio-temporal context learning method with saliency feature which obtained by improved saliency detection algorithm.Secondly,the size of ROI(Regions of Interest)may occur changes which caused by camera movement or the different distance between camera and target.Therefore,this paper proposes a multi-scale and multi-feature tracking algorithm based on hash function.We integrate multiple cues for target representation by constructed hash function.We predict target locationby an online weighted instance learning classifier in next frame,and estimate the size of target by a additional scale filter.Finally,we applied tracking algorithm to practical application,and proposed a loitering detection algorithm by analyzing the trajectory of pedestrian.We judge the loitering behavior by analyzing the trajectory of object through the duration time and the trajectory’s angles variation.Compared with the method which only using the duration time or the confusion degree of trajectory,our method provides superior performance in terms of detection rate and false alarm rate. |
