3-D Object Reconstruction From Outdoor Ultrasonic Image and Variation Autoencoder

We present a technique for three-dimensional (3D) object reconstruction utilizing an ultrasonic array sensor and a variational autoencoder (VAE) within a high-interference environment. In scenarios where optical instruments such as cameras and LiDAR are impractical, the utilization of air-coupled ultrasonic waves for 3D measurements has emerged as a viable alternative. Nevertheless, deploying this technology in high-interference settings, particularly outdoor environments, has presented significant challenges. To tackle this challenge, we have developed and established a methodology for the 3D reconstruction of stationary objects by combining the time-of-flight point cloud data acquired through beamforming with the deep learning model VAE. This study proceeds by elucidating the procedure for conducting beamforming and measuring distances using ultrasonic waves. Subsequently, we expound upon an experimental methodology that employs 3D object reconstruction and associated techniques. Finally, we present the results obtained from attaching an ultrasonic sensor to a utility pole and conducting ultrasonic measurements. Our investigation focuses on four distinct types of objects: boxes, motorbikes, humans, and reflectors. The measurement system was positioned 5 m above the ground on a utility pole situated alongside the road. The objects selected for measurement were situated in stationary positions within a $3~\text {m}^{{3}}$ area, with a maximum distance of 10 m from the utility pole. The objective of this study is to assess the efficacy of ultrasonic measurements and object reconstruction techniques under these specified conditions. The results indicate a precision of 0.939, a recall of 0.868, and an F-value of 0.902, which are considered sufficient for the application of ultrasonic waves.