Description

Locating targets play an important role in nature. Some species may have specific anatomic characteristics that allow enhancing accuracy when finding their prey. In particular, barn owls have an asymmetric ears disposition that made them outstanding night hunters. On the other hand, this capacity is also relevant to engineering applications: radars, GPS and sonars are great examples. When it comes to sound source localization (SSL) systems, the time difference of arrival (TDOA) estimation between sensors is a very common approach. This method is very dependant on three factors: the geometry of the sensors' disposition, the synchronization between sensors and the sampling rate of the received signals. iOwlT is an intelligent cyber-physical SSL system inspired by owls and focused on locating impulsive sounds, as gunshots. In this work, the system was physically built and trained to recognize and locate some kinds of impulsive sounds. Also, it presents a comprehensive simulation study on optimizing SSL hardware parameters with particle swarm optimization (PSO) algorithm. The system's sound classification, based on neural networks, scored 91.38% accuracy, and the localization algorithm performed 97.21% and 88.32% on determining impulsive sounds direction and position, respectively. The optimization method presented potential directions on building better SSL systems, performing up to 33.0% better than similar problem approaches.

Key-words: cyber-physical systems, sound source localization, particle swarm optimization, biologically-inspired systems, sensor array.

Submission
Poster
Presentation
Final Report
Senior Thesis