基于机器学习的无人机载LiDAR数据处理系统研究

Research on UAV LiDAR data processing system based on machine learning

为了进一步减少无人机载激光雷达传输给飞行控制器的数据量,对已有的一种数据预处理系统进行改进。无人机避障过程中,用于测量障碍物信息的激光雷达数据中存在着无效、冗余等数据,这些多余数据给无人机飞行控制器带来巨大的数据处理负担。在已有的数据处理系统的基础上,改用基于密度的具有噪声的聚类方法(DBSCAN)辨认障碍物,把测量为连续线性的点识别为一个物体。实验证明,改进系统不但大幅减少了传送给飞行控制器的数据量,而且构建出激光雷达原数据中不存在的障碍物的宽度信息,还将辨识障碍物的准确率提高多达53%。

In order to further reduce the amount of data transmitted to flight controller by UAV-borne lidar,an existing data preprocessing system is improved.In the process of UAV obstacle avoidance,there are invalid and redundant data in the lidar data used to measure obstacle information,and these redundant data bring huge data processing burden to UAV flight controller.Based on the existing data processing system,the DBSCAN clustering machine algorithm is used to identify obstacles,and the points measured as continuous linear are recognized as an object.Experiments show that the improved system not only greatly reduces the amount of data transmitted to the flight controller,but also constructs the width information of obstacles that do not exist in the original data of lidar,and improves the accuracy of obstacle identification by up to 53%.