Neural network-based geometric references recognition applied to ultrasound echo signals

Abstract

Ultrasonic sensing systems are often used in robotics for navigation purposes, such as obstacle avoidance and distance measurements. However, a very desirable but difficult task is the detection and recognition of geometric references. This is a tough task due to the significant interference caused to the ultrasonic sensing system by the environment, such as temperature variations and air convection or wind. Furthermore, the echo signal variation due to the various detected references is nonlinear. The use of neural networks is strongly recommended when the process involved is nonlinear or has a very complex mathematical model, which is exactly the present case. The application of neural networks to the recognition of geometric references via ultrasound obtained very good results.