Night-Time Measurement and Skeleton Recognition Using Unmanned Aerial Vehicles Equipped With LiDAR Sensors Based on Deep-Learning Algorithms

Abstract

Robots, unmanned vehicles, and unmanned aerial vehicles (UAVs) are widely used as mobile platforms to facilitate the scanning range and coordinate positioning of vision and recognition for analyzing surrounding environmental information. However, during the night, image recognition is not efficient. In this study, a UAV equipped with a LiDAR sensor was used to detect night-time environmental information and recognize human characteristics. The point data of the foreground and background were separated using the background difference method and density-based spatial clustering of applications with noise (DBSCAN) to automatically distinguish and classify objects by the distance difference between adjacent points. Root mean square error (RMSE) was calculated by surface fitting, and the surface state of the object was summarized as a condition to determine whether the object is human. Image processing was performed on point cloud data, and skeleton recognition was performed using the surface image of the object and artificial intelligence (AI) algorithm. Thus, an automatic recognition system was established. Hyperparameters were adjusted for optimal modeling. Finally, VGG16 was used for feature extraction to obtain a mean average precision (mAP) of 95.3%. Human body recognition using skeleton recognition based on RMSE surface analysis approached 93.8% during the night. The website information of recognition results was established through the server terminal computer to present the current measured environmental state and recognize a human body in the night-time environment. This technology can be used to search for victims during disaster relief scenarios and reduce the difficulty and time required for night-time search and rescue operations.