Stereo Vision And Lidar Data Fusion

3D Lidar(Light Detection and Ranging) is an advanced technology for rapid, accurate and non-contact acquisition of spatial location information of real objects, providing source data for large-scale objects or those with complicated geometric shape, and thus has shown huge potential application prospects. However, acquired data is hard to be complete for the reasons like Lidar's position, occlusion and so on. Completing lost data is often costly or even impossible by the same way, which leads to incomplete reconstructed models with part of surface or details missing. Stereo vision, another kind of important 3D reconstruction technology, has a distinct advantage in convenience, but disadvantage in efficiency. Most studies only focus on one of the two kinds of reconstruction technologies. In this paper, the complementarity of the two technologies in data acquisition and data processing is applied to recovery of lost data by fusing the data of stereo vision and Lidar.The main work of this paper includes:(1) The principles, methods and technical features of both Lidar and stereo vision are discussed. The analysis of their strengths and weaknesses is given as a basis for the necessity and possibility of their combination, which provides a basis for presenting the scheme of completing Lidar data by fusing 3D data calculated based on stereo vision with Lidar data.(2) The different characteristics of the image matching algorithm are researched, and the phase correlation matching algorithm is improved to avoid the influence of the image rotation. Firstly, image feature matching is carried out; secondly, epipolar correction is executed; at last, along the scan line, the dense phase correlation matching is realized. Through these three steps,the false matching rate is reduced.(3) The fusion of the data of Lidar and stereo vision is achieved, further more, to improve the accuracy of the fusion, Euclid vector in the algorithm of ICP(Iterative Closest Point)including the scale factor is corrected according to the normal vector of Lidar data.The above algorithms are realized on the platform VC++ , OpenCV, and OpenGL, tested with the data acquired by our Lidar instrument and high-resolution camera, and finally proved valid by experimental results.