Scene Change Detection For Unmanned Aerial Vehicle Cruise Application

Recently,it is commonly to use machines to replace humans to repeat repetitive tedious or dangerous work in many areas.The use of unmanned aerial vehicles(UAV)is a very typical example.With the increasing application of UAV in military reconnaissance,resource detection and disaster early warning,the research on the automatic processing of UAV images is also deepening.In the process of UAV cruise video,due to UAV flight routes and flight angle changes and other reasons,there are many changes in the scene.In order to extract the information in the UAV cruise video more accurately and effectively,we need to detect the scene changes in the video.In this thesis we studied the visual inspection method of scene change under UAV cruise application,and proposed a scene change detection method based on spatiotemporal significance and a scene change detection method based on generalized Gaussian mixture model.The scene changes in the video sequence are similar to the motion of the target.So it's possible to apply the relevant method of moving target detection to the scene change detection.In this thesis,the method of applying the spatiotemporal significance and the generalized Gaussian mixture model to the scene change detection is studied deeply.The main contents of this thesis are as follows:(1)Based on the traditional scene change detection algorithm,we studied the scene change detection method based on motion target detection.And the principle and realization of the relevant detection method are briefly described.The feasibility of applying the relevant method in motion target detection to scene change detection is analyzed.(2)On the basis of image significance detection,we studied the spatiotemporal significance detection.The spatiotemporal significance contains both of the time information and spatial information of the image sequence.This paper combines the color and position information to calculate the spatial characteristics,and then use the low rank sparse decomposition to get motion information.And the spatiotemporal significance is combined with the traditional three-frame difference method to extract the scene changes in the video sequence.Experiments show that the algorithm can effectively detect the area where the scene changes in the video.(3)Based on the similarity of scene change detection and motion target detection,we studied the background modeling algorithm based on Gaussian mixture model.The Gaussian mixture model is extended to the generalized Gaussian mixture model,and the generalized Gaussian mixture model is applied to the scene change detection.The scene change detection algorithm based on the generalized Gaussian mixture model can effectively detect the area where the scene changes and the detection effect is superior to the Gaussian mixture model method.