Laser And Vision Data Fusion For3D Scene Reconstruction And Surveillance

3D scene reconstruction and surveillance based on vision is sensitive to environmental factors such as the light changes, and can’t provide location information of the scene directly. So there are some limitations in practical applications.3D laser can obtain the range information and is not sensitive to the environmental factors. So how to fuse the3D laser range with vision data and apply to both two files is becoming a hot area of research.The precise calibration of extrinsic parameters between3D laser and vision is the basis of data fusion. In order to calibrate the extrinsic parameters, a novel chessboard calibration board with round holes and red marking band is designed. Matching laser and vision corners are need to calibrate the extrinsic parameters. In order to extract the3D laser corners, the ICP (Iterative Closest Point) algorithm is conducted based on the initial result to improve the accuracy of corners. In order to get vision corners, the adaptive threshold is applied. Verification experiment and quantitative analysis show the effectiveness and feasibility of the calibration method proposed in this paper.In order to reconstruct the scene effectively, an effective triangular meshing method based on the point cloud order is adopted for3D scene modeling. Meanwhile a data fusion approach based on laser and vision triangular mesh is proposed for scene color and texture display. A local outlier detection algorithm and restoration process is applied to remove the outliers as well as keep the order. Then build triangular mesh model of the point cloud and remove the error divisioa Each triangular mesh surface is mapped with texture by the extrinsic parameters. Finally scene reconstruction experiments show the scenes reconstructed have the same geometry and texture information with real scene.Traditional surveillance method based on monocular vision for surveillance with multi-objective occlusion has certain limitations due to lack of range information. In this paper human target is detected by laser elevation map, and the target points are mapped to images by the mapping relationship for multi-target detection and display. Experiments show the effectiveness of the method proposed in paper.