一种新的尖轨点云配准方法及算法加速研究

A Novel Point Cloud Registration Method of Tongue Rail and Algorithm Acceleration

为满足高速道岔检测的严格要求,解决尖轨检测中三维点云配准定位难的问题,提出利用尖轨轨腰圆孔特征配准的方法。首先在轨腰上制造圆孔特征,再提取圆孔边界作为特征初始匹配,最后以轨腰作为基准精确匹配。针对边界提取算法中运算效率低下的问题,提出基于异构计算语言的边界提取并行算法。通过分析数据和算法特点,改进了点云数据的存储方式并修改了部分算法以提高计算效率。试验表明,提出的方法有效解决了尖轨配准中定位问题并满足检测标准;改进的算法相对于原算法提高了计算效率,并保证了和系统良好的兼容性。

A registration method of tongue rail based on waist hole feature is proposed to meet the stringent requirements of high speed railway turnout detection and solve the difficult problem of 3D registration for tongue rail detection. This novel method created hole features in the rail waist and extracted the boundary of the circular hole as the initial matching. The rail waist was used as the accurate registration. In order to solve the problem of low efficiency in the boundary extraction algorithm, we proposed a new parallel algorithm of boundary extraction based on heterogeneous computing. By analyzing the data and calculation characteristics, the storage mode of the point data were improved and some of the algorithms was modified to improve the computational efficiency. The experimental results show that this method can effectively solve the problem of registration for tongue rail and meet the testing standards; the proposed algorithm can improve the efficiency compared with the original algorithm and ensure good compatibility with the system.