Position and posture measurements and shape recognition of columnar objects using an ultrasonic sensor array and neural networks

Abstract

SUMMARYA new method for measuring the position and posture (pose) and recognizing the shape using ultrasonic wave sound pressure signals and neural networks is proposed. In most of the past methods of this kind, the characteristic quantities have been based on either time of flight methods or acoustic holographic methods. In these methods, measuring and recognizing the width and depth directions simultaneously with a high resolution has been difficult in principle. In this paper, the problems of the past methods are resolved by changing the viewpoints by using criteria such as the peak values of the penetrating and reflecting waveforms, their positions, their deflection points, and their intervals when ultrasonic waves are directed onto a measured object, as characteristic quantities given to neural networks. In addition, a prototype measuring and recognizing system implementing the proposed scheme is constructed by introducing a circuit for detecting the peak positions of the sound pressure signal distribution waveforms with high resolution. The experimental results are satisfactory in position measurements and shape recognition, and confirm that the method is applicable to rectangular prisms and the like. It is shown that the position and pose measurements and shape recognition can be realized with high resolution in both the width and depth directions by this scheme, and that the proposed method is very effective when applied to position and pose measurements and shape recognition and the like ultrasonically.

Category recognition system using two ultrasonic sensors and combinational logic circuit

“…Usually, the information obtained from the reflected waves by the ultrasonic sensors is only the presence or absence of an obstacle in the measurement range or the distance information to the target object. However, research has been performed on the recognition of the shapes of objects by using neural networks from the information obtained from ultrasonic sensors [7][8][9]. In the system in Ref.…”