基于灰度相似性的激光点云与全景影像配准

Registration of laser point cloud and panoramic image based on gray similarity

针对传感器参数未知、环境结构特征不明显、影像数据较少等情况,提出基于灰度相似性的车载3D激光点云与全景影像自动化配准方法.基于全景拼接算法和柱面投影原理,分别将多张单幅图片拼接为全景影像,并将3D激光点云转换为2D深度图.基于灰度相似性原理,将全景影像和2D深度图在水平方向和垂直方向等间隔细分成区域对,沿水平方向和垂直方向移动全景影像,计算每次移动后各细分区域对之间像素灰度值之和的比值,并求解其均方差,将均方差最小时的区域移动值作为最终匹配偏移量.根据偏移量计算得到全景影像相对3D激光点云的水平旋转角度和垂直平移距离.实测结果表明,本研究所提算法对场景的适应性较好,平均配准误差为2个像素,而对比方法无法实现有效配准.

An automatic registration between vehicle 3D laser point cloud and panoramic image based on gray similarity was proposed,under the scenes of unknown sensor parameters,unclear environmental structure characteristics and small amount of image data.Firstly,multiple single images were spliced into panoramic image and 3D laser point cloud was converted into 2D depth image,respectively,based on the panoramic stitching algorithm and cylindrical projection principle.Secondly,based on the principle of gray similarity,the panoramic image and 2D depth map were subdivided into region pairs at equal intervals along the horizontal and vertical directions,and the panoramic image was moved along the horizontal and vertical directions.The proportion of the sum of pixel gray values between each pair of subdivided regions after each move was calculated,and its mean square deviation was solved,and the region move value with the smallest mean square deviation was taken as the final matching offset.Finally,the horizontal rotation angle and vertical translation distance of the panoramic image relative to the 3D laser point cloud were calculated according to the offsets.Experimental results show that the algorithm has good adaptability to the scenes and the average registration error was 2 pixels,while the comparison method cannot achieve effective registration.