Automatic registration framework for multi-platform point cloud data in natural forests

The use of light detection and ranging (LiDAR) for investigating forest parameters has gained attention in recent years. However, the occlusion of trees in natural forests makes it difficult for LiDAR on a single platform to capture complete point clouds of trees. In order to solve this problem, it is crucial to combine multi-platform LiDAR data. Because of the complexity of natural forests and the small difference between the geometric characteristics of trees, current multi-platform LiDAR data fusion remains an ongoing challenge in natural forests. In this paper, an automatic registration framework for multi-platform point cloud data in natural forests based on tree distribution pattern was proposed. It consists of five steps, namely segmenting trees, generating feature descriptors, matching trees and registering coarsely and finely. The proposed registration framework can determine the same and accurate location information of matching trees from multi-platform LiDAR data, and a large number of correct matching trees can be obtained through two rounds of a single tree matching process. The proposed framework was validated by fusing airborne laser scanner (ALS) and backpack laser scanner (BLS) data in natural forest. According to experimental results, the proposed framework has a high registration accuracy (root-mean-square error (RMSE) = 0.133 m, mean absolute error (MAE) = 0.126 m). In addition, when the single tree segmentation accuracy exceeds 0.85, the proposed framework is less affected by segmentation errors. In natural forests, the proposed framework can effectively improve the accuracy and efficiency of multi-platform LiDAR data registration.