Dense Depthmap Prediction from Ultrasonic Sensors

Koki Yasui, Fumihiko Sakaue, Jun Sato, Yu Koyama, Mitsuyasu Matsuura

Abstract: In this paper, we propose a method for estimating a dense depth map of a scene by combining information acquired from multiple ultrasonic sensors. In this method, the probabilities of the presence of an object at each discretized point in the scene are estimated based on the observed signals, and the depth map is estimated from this probability map. We also show that the depth map estimation can be stabilized by introducing prior knowledge such as the sparsity of the 3D scene as a regularization. Experiments on real and synthetic scenes confirm that the proposed method can estimate the depth map.