基于自适应双阈值的车窗三维点云修复算法-《计算机技术与发展》

文章信息/Info

Title:: 3D Point Cloud Data Repair Algorithm for Car Window Based onAdaptive Dual Threshold

Author(s):: TAN Ke; MA Rong-gui; LUO Lei; School of Information Engineering,Chang’an University,Xi’an 710061,China

Keywords:: point cloud processing; boundary extraction; hole repair; dual-threshold; radial basis function

摘要:: 目前车辆轮廓检测主要是使用激光扫描技术,能够快速获取复杂曲面的点云数据。在扫描车窗处时,因为车窗本身玻璃材质的透光率较大,产生异常噪声点,污染车窗及其周围的数据,影响车辆轮廓检测的精度。针对车窗处点云数据异常的问题,提出基于点间距与曲率的自适应双阈值特征提取算法。首先计算车辆切面点间距变化量与曲率,然后基于标准差确定双阈值,获得车窗噪声特征点边界。根据特征点边界对车窗部分的边界进行拟合,去除边界内的噪声点。再根据车窗点云数据的特征选择合适的径向基函数对车窗部分进行曲面确定,最后对曲面进行插值完成孔洞修复。通过对激光雷达检测到的车辆三维点云数据进行车窗数据修复,对算法进行检验,结果表明该算法可对车窗数据进行修复且效果良好。

Abstract:: Currently,vehicle contour detection mainly uses three-dimensional optical scanning technology,which can quickly obtain the point cloud data of complex curved surface. When scanning the car window,as window glass normally has large transmittance,abnormal noisy points will be generated,which compromised the data of the window and its surrounding area and in turn affected the accuracy of vehicle contour detection. An adaptive double-threshold denoising algorithm based on point spacing and curvature is proposed to handle with target abnormal point cloud data at car window. Firstly,the distance variation and curvature of vehicle tangent point are calculated,and the double threshold is determined based on standard variation,so as to obtain the boundary of window noise characteristic points.The boundary of the window is then fitted according to the boundary of the characteristic point to remove any noisy points in the boundary. Secondly,based on characteristics of the window point cloud data,the appropriate radial basis function is selected to interpolate window surface to complete the hole repair. The proposed algorithm is verified by repairing the window data of the vehicle data detected by LiDAR,and the results have demonstrated that it can repair window data with better effect.