基于激光SLAM的小麦点云采集系统与冠层高度提取方法

Point Cloud Acquisition and Canopy Geometric Features in Wheat Based on Laser SLAM

摘要:

为了能够提高田间作物三维信息获取的准确性与效率，以小麦为研究对象，开发了一套田间多传感器数据采集装置，以自走式车辆为移动载体，利用三轴云台作为增稳载体，构建了一套激光雷达和IMU紧耦合点云采集系统。通过研究传感器的成像特性和采集方式，提出了一种基于激光SLAM的采集方法来构建田间高精度点云地图，从而准确获取田间作物点云信息，能够以1.5 m/s的速度完成地图构建，不需要额外增加田间标靶，节约了后期点云匹配的资源。在点云地图的基础上，使用直通滤波、基于Octree的下采样和统计滤波完成了前处理。提出一种基于垂直度和高度模型的地面区域精准提取方法，针对小麦生长期间根部点云难以获取，使用点云PCA分析计算点云法向量进行垂直度提取，经过二次结合高度模型成功分割出不规则的地面点，再次利用地面稳定拟合平面计算新的冠层高度模型。通过统计分析，与人工测量真值相比，基于SLAM的田间小麦三维地图，其建图精度均方根误差可以达到0.04 m;同时本文的冠层高度提取算法与人工测量真值相关系数达到了0.979。研究可以为小麦田间三维性状采集系统设计和性状分析提供有力工具。

关键词:小麦;激光SLAM;三维点云;信息采集;冠层高度

Abstract:

In order to be able to improve the accuracy and efficiency of the acquisition of three-dimensional information of field crops, taking wheat as the research object, this paper develops a set of field multi-sensor data acquisition device, using a self-propelled vehicle as the mobile carrier and a three-axis gimbal as the stabilisation carrier, and a tightly coupled point cloud acquisition system of LiDAR and IMU was constructed. By studying the imaging characteristics of the sensors and the acquisition method, a laser SLAM-based acquisition method was proposed to construct a high-precision point cloud map in the field, so as to accurately acquire the point cloud information of crops in the field, and be able to complete the construction of the map at a speed of 1.5 m/s, without the need to add additional field targets, which saved the resources for matching the point cloud at a later stage. On the basis of the point cloud map, pre-processing was completed by using straight through filtering, Octree-based downsampling and statistical filtering. An accurate extraction method of ground area based on verticality and height model was proposed. For the difficulty of obtaining the root point cloud during the growth period of wheat, the point cloud PCA analysis was used to calculate the normal vector of the point cloud for the calculation of verticality, and the secondary combination of the height model successfully segmented out the irregular ground points, and the new canopy height model was calculated by using the ground stabilisation fitting plane. Through statistical analysis, compared with the true value of manual measurement, the accuracy of SLAM-based three-dimensional map of wheat in the field, the root mean square error can reach 0.04 m;at the same time, the correlation coefficient between the canopy height extraction algorithm and the true value of manual measurement reached 0.979. The research can provide a powerful tool for the design of the three-dimensional trait collection system and trait analysis of wheat in the field.

Key words:wheat;laser SLAM;3D point cloud;information acquisition;canopy height