Research Of Dynamic Target Detection And Behavior Recognition Method Based On Geographic Constraint Scene

The methods of traditional dynamic target detection based on surveillance video,and relate behavior analysis work after it,mostly focus on the video images,such as the human body structure and image color similarity.These methods ignore the large amount of geographic information contained of surveillance video,which are difficult to associate video monitor with geospatial information.The analysis and understanding ability of video monitoring information improve by integrating the video monitoring information with the spatiotemporal information of geospatial space,and giving dynamic attribute to the static geospatial information.Therefore,in this paper,the video image information and geographic space information are mutual mapping by establishing a mutual mapping model of video monitor images and geographic space.Besides,the problems of indoor dynamic target positioning and dynamic target behavior recognition are solve by the method of researching spatial feature information extraction of indoor dynamic target,and behavior recognition works.Using these methods can integrate organic integration of video monitor with geospatial information.The content and results of this article are as follows:(1)A geospatial coordinate mapping model of video monitor images based on EPnP is established.Endowing the image pixels to the geographic space coordinates by finding the mutual correspondence between the coordinates of the image pixels and the corresponding points in the geographic space.Then use this model to achieve the precise positioning of the indoor dynamic target contained in the video frame in the geographic space,that achieve the fast and accurate positioning of the dynamic target.(2)A method of moving object detection by combining frame difference with improved Vibe are proposed.By adding the detection of scintillation pixels and combining the shadow removal method of dynamic target detection based on HSV color space,the problem of ghosts and shadows generated effectively solved by Vibe algorithm processing dynamic targets,and a complete foreground target image is obtained.(3)A method of extracting the information of spatio-temporal features of dynamic targets in geographic constraint space is design.This method extracts the relevant features of the dynamic target in each single frame image originally,and then uses the geographic space coordinate mutual mapping model of the video monitor image based on EPnP to calculate the geographical space coordinates of the single frame target.In addition,the perceptual hash algorithm is used to realize the stable tracking of the dynamic target of the pedestrians in the video frame sequence,and then to obtain the spatio-temporal feature information like speed,acceleration,and direction of motion.Finally,realized the spatial and temporal feature data extraction of the dynamic target of video continuous frame sequence.(4)Researching a method of dynamic target behavior detection and recognition based on SVM.This method extracts the centroid of the dynamic target,calculates its minimum circumscribed rectangle aspect ratio,normalized centroid M1 value of the dynamic target and the change rate of the dynamic target centroid in the image coordinate system u and v directions.Finally realized the detection and recognition of multiple dynamic target behaviors by constructing a classifier and using SVM for training.