基于Scan Context与NDT-ICP相融合的果园建图方法研究

由于果园环境复杂、树冠特征单一，且叶片会引起光线漫反射等因素，导致果园环境的地图构建过程中出现误匹配，增大建图的累计误差。针对以上问题提出一种基于Scan Context与NDT-ICP相融合的果园环境导航地图构建方法。该方法首先对Ring key进行快速地上层搜索，得到候选帧，并对候选帧与当前帧进行相似度评分，通过两阶段搜索算法来有效地检测回环以减少果园环境地图中的误匹配。同时使用基于正态分布变换粗配准与迭代最近点精确配准融合的点云配准方法降低果园环境地图的累计误差。试验结果表明，在KITTI数据集中使用半径搜索回环检测的回环数为42，使用Scan Context回环检测的回环数为51，回环数提高21.4%。NDT-ICP匹配算法的均方根误差为12.86 m，ICP匹配算法的均方根误差为15.11 m。在真实果园环境中使用半径搜索回环检测的回环数为196，使用Scan Context回环检测的回环数为261，回环数提高33.2%。该研究算法有效地降低果园环境地图构建过程中的误匹配、累计误差大的影响，满足果园环境下的高精度环境建图需求，为推进果园环境的无人化作业提供技术支撑。

Research on orchard SLAM method based on Scan Context and NDT-ICP fusion

Due to the complex orchard environment, single canopy features, and the diffuse reflection of light caused by leaves, the mapping process of orchard environment is mismatched, which increases the cumulative error of mapping. In order to solve the above problems, a method of constructing orchard environment navigation map based on the fusion of Scan Context and NDT-ICP was proposed. In this method, the Ring key is searched quickly to obtain candidate frames, and the similarity between candidate frames and the current frame is scored. The two stage search algorithm is used to effectively detect the loop and reduce the mismatching in the orchard environment map. At the same time, the point cloud registration method based on the fusion of normal distribution transformation rough registration and iterative nearest point accurate registration was used to reduce the cumulative error of the orchard environmental map. The test results show that in KITTI data set, the loopback number of radius search loopback detection method is 42, and the loopback number of Scan Context loopback detection method is 51, improving the loopback number by 21.4%. The root mean square error of NDT-ICP matching algorithm is 12.86 m, and that of ICP matching algorithm is 15.11 m. In the real orchard environment, the number of loop detection using radius search is 196, and the number of loop detection using Scan Context is 261, which increases by 33.2%. The algorithm can effectively reduce the influence of mismatching and large cumulative error in the process of orchard environmental mapping, meet the demand of high precision environmental map construction in the orchard environment, and provide technical support for promoting the unmanned operation of the orchard environment.