Surface-Condition Detection System of Drone-Landing Space using Ultrasonic Waves and Deep Learning

We propose a system for detecting the surface conditions of a landing space using ultrasonic sensors mounted on a drone. The advantage of ultrasonic sensors is that they are extremely low cost, are much lighter and smaller than cameras, have millimeter-wave lasers, and use LiDAR. However, normal ultrasonic sensors can only measure the distance from the nearest object, so the amount of information is insufficient to estimate the conditions of a landing space. Therefore, we propose installing an ultrasonic sensor on each arm of the drone and estimating the condition of the landing space from the time series of reflected waves for very short ultrasonic waves. In the measurement results, reflected waves were small and changed irregularly for each sensor where a space was not suitable for landing. In a simulation experiment using deep learning, our system was able to determine whether a condition was suitable for landing with an accuracy of 98%.