Mapeamento de objetos em sonares utilizando compressed sensing para navegação autônoma

Abstract

Sonars are widely used in navigation systems, which traditionally use the time-of-flight technique for obstacle detection and localization. This technique computes the time between sending and receiving the reflected ultrasonic signals. However, this technique presents problems with multiple ultrasonic reflection interferences in environments with multiple obstacles, requiring several acquisitions for an appropriate detection. This dissertation presents a novel approach to obstacle detection and localization. Some characteristics inherent to the ultrasound signals in the time domain, inverse problems, and Compressed Sensing concepts were used. Experiments were conducted in a controlled environment with multiple obstacles, and the ultrasonic reflections were acquired by a hardware platform with four sonars explicitly developed for sending and acquiring raw ultrasonic signals. The proposed approach was validated by comparing the reconstructed images – a bi-dimensional occupancy grid format – using the Orthogonal Matching Pursuit algorithm, a traditional image reconstruction method, and Tikhonov regularization. Three different representations of signal values were used, and the reconstructions were evaluated quantitatively, with the Compressed Sensing method showing promising results, which recommend its use in research for the detection and location of obstacles using sonar in the air.