Depth Sensor Application in Ground Unevenness Estimation for UAV Emergency Landing

As research on unmanned aerial vehicles (UAVs) progresses rapidly, so does research on emergency landings. In recent years, with the increasing demand in urban environments and disaster areas, there have been attempts to land on flat surfaces, but there is no way to land on uneven ground. Therefore, we utilize an on-UAV depth camera to estimate the unevenness of the ground. However, directly using depth information to estimate ground unevenness results in excessive computational complexity. Additionally, the presence of noise disrupts the accuracy of distance calculations. Due to these factors, it is necessary to reduce computational costs and extract ground unevenness features unaffected by noise. In this research, we propose a new method for estimating the degree of unevenness of the ground using machine learning and statistical processing methods for depth information obtained from the camera. After applying the newly proposed cell-based data segmentation to the depth information and performing segmentation and feature extraction of the ground using Principal Component Analysis (PCA), Mahalanobis' distance is utilized to evaluate the unevenness of the ground. By appropriately adjusting the parameter values for the degree of unevenness to be estimated according to the adaptation environment, it is possible to estimate the uneven area that does not interfere with the landing of the UAV from a flat surface. We confirmed the effectiveness of the proposed method in a verification experiment. The proposed method can achieve stable and highly accurate estimation of ground unevenness.