| The conception of next-generation Digital Earth calls for that Digital Earth System should not only concern with the affinitive data from GIS and Remote Sensing,but also parse,visualize,and analyze some real-time and dynamic data obtained from Internet of Things.As an important component of senser network,real-time videos are the most commonly available and location-related data that can carry content-rich information and provide dynamic scenes of the real world at present or in the past.Fusing real-time videos with virtual scenes in Digital Earth system to establish an Augmented Virtual Environment can take full advantage of spatial localization and visualization of the 3D scene,and significantly extend the applications of video processing results.Thus it will have important significance for some time-critical applications,such as intelligent transportation,public security issues,3D video surveillance,emergency response and management,urban planning and management,location and navigation service.Combining real-time surveillance videos and Digital Earth Science Platform,a series of methods for 3D simulation and analysis of moving objects were proposed,including real-time moving objects tracking and 3D visualization,3D path analysis,intelligent motion perception,possible multi-paths prediction and optimal path planning.This dissertation aims to integrate the technologies of computer vison and Digital Earth,promote the development of Digital Earth in time-critical applications,provide technical support and aid decision making of intelligent 3D video surveillance and emergency response.The main contributions of this dissertation are illustrated as follows.1.A real-time video object tracking method was developed to obtain clear silhouettes and pixel coordinates of the objects efficiently.This real-time video object tracking method is composed of a multi-thresholds improved Gauss Mixed Model background difference algorithm and a foreground optimization algorithm based on HOG featuresa and SVM classifier.The test results showed that the real-time video object tracking method only took about 58 ms to track and extract objects from a video frame so that it could provide real-time,accurate and nessary data for 3D visualization,movement simulation and analysis of moving objects.2.A new 3D visualization method for moving objects based on complex 3D scene and virtual projection was proposed.Combining the results of object tracking method,complex virtual scene and the perspective projection relationship between virtual projector and 3D scene,a 3D visualization method for moving objects based on complex 3D scene and virtual projection was proposed.This method included 3D modelling of objects,3D location of object model,texture projection and 3D trajectory visualization.This method can run a 3D scene containing 144 thousand triangles and a moving object at around 50 frames per second on a laptop with 1GB of graphics card memory,which means the method can guarantee the smooth visualization of a moving object in a complex virtual scene.3.Several 3D path analysis algorithms were discussed and a new quaternion-based section-by-section 3D modelling method for path network was proposed.The indoor/outdoor paths were classified according to their types,and abstracted into topological network with nodes and edges.The Dijkstra algorithm,A* algorithm,bidirectional Dijkstra algrothm and bidirectional A* algorithm were introduced and a height heuristic A* algorithm for 3D path analysis was proposed in this dissertation.The test results showed that all the five algorithms could find the optimal path quickly and accurately,taking lessthan 2ms to find the shortest path from a network with 500 egdges.A quaternion-based section-by-section 3D modelling method for path network was proposed to automatically generate vivide and highly recognizable 3D models for the path network.4.A new approach for moving objects simulation and path routing in a 3D scene with multiple surveillance cameras was proposed according to the demands of 3D video surveillance and emergency evacuation.A multi-cameras cooperative perception and visualization algorithm is proposed to re-recongnize the moving objects,visualize their trajectories and attribute infoemation in virtual scene.The test results showed that it could run a 3D scene containing 295 thousand triangles at 35.7 fps.A possible multi-paths prediction algorithm is proposed to estimate the motion direction of a specific target,predict the possible paths which the target would choose,visualize the possible multi-paths with different colour and dispatch the surveillance cameras intelligently to track the target continuously.Then an optimal path planning algorithm for emergency evacuation was proposed to detect the trapped persons in a building,find the optimal evacuation route and provide accurate evacuation guidance as soon as possible. |
PDF Full Text Request