Maize Point Cloud Data Filtering Algorithm Based on Vehicle 3D LiDAR

Abstract

In order to support phenotypic parameter measurement and digital plant related research, the obtained maize point cloud data collected by 3D light detection and ranging (LiDAR) were analyzed and processed. The filtering algorithm of maize point cloud data was carried out, and a two times filtering algorithm based on statistical analysis was proposed. The vegetative stages of the 12th leaf, Jingnongke 728 and Nongda 84 maize were used as research objects, and VLP-16 was used to collect field maize point cloud data. Firstly, the point cloud data was subjected to pass filtering processing to remove extraneous points. The number of point clouds was reduced from 12 000 to 1 700. Secondly, the point cloud data was subjected to the first filtered process, and the precision and recall threshold were set. The average number of point clouds was reduced from 1 700 to 1 400, and 300 outliers were removed. Then, the point cloud was subjected to the second filtered process. The optimal combination and marginal combinations of precision and recall were determined. The optimal combination was (110, 0.9) and (6, 1.2). The marginal combinations were (100, 1.0), (6, 1.2) and (110, 0.8), (5, 0.9), a total of three combinations of parameters. The average number of point clouds was reduced from 1 400 to 1 300, and 100 outliers were removed. Finally, the three sets of verification set data were tested. The results showed that the optimal combination performance was optimal, which can be used to Jingnongke 728 and Nongda 84.