Research Of Traffic Surveillance Algorithm Based On Two-view Stereo Vision

At present, Most of the traffic surveillance nodes are single-view monitoring system.The fact that the traffic target could not be accurate located by the system causes someinconvenience in the traffic surveillance. Furthermore it can not meet the requirements ofintelligent transportation. But the two-view stereo vision will be able to make up theweakness. The three-dimensional space data of the road scene can be calculated, and thento locate the positioning of the object. It will greatly enhance the intelligence andautomation of the traffic surveillance. This paper focus on research the basic technologyof the binocular stereo vision, and build the specific application for traffic surveillancestep by step. The main investigative contents of this paper include the following:1. General study on the technical knowledge of two-view stereo vision, focus oncamera calibration and3D reconstruction. Bundle the algorithms and the applications toresearch, develop the most suitable stereo algorithms for traffic surveillance.2. The computer simulation and real camera combined to verify the experimentsthrough the whole paper. Build the virtual road scene in3ds Max to obtain the variousmaterial easily, can be improve the efficiency of the algorithm validation. Furthermore,introducing the simulation to verify the algorithms of stereo vision, for example,including but not limited to the camera calibration, Rodrigues rotation formula, sub-pixelcalculation,3D reconstruction. 3. Studied the binocular camera calibration method in mathematical way. Not onlysatisfy to obtain the final data, more focus on the correlation among the parameters.Expand the application field, proposed the convergent point positioning methods fordual-view stereo camera.4. Study on the vehicles extraction method under road scene to improve theautomation of binocular stereo monitoring. Modularize the detection method into someparts, like extract foreground, background real-time update, shadow detection andremoval, to enhance the optimization of algorithm in specified scene.5. Proposed a new3D reconstruction algorithm based on relative extrinsicparameters to achieve a best integrity with versatility, computational speed and accuracy.