A measurement setup for localising people in indoor environment based on a system characterized by acquisition of audio files and images implemented on a sensorised social robot is proposed. The audio signal processing for human voice identification applies the segmentation methodology for the direction of arrival (DOA) estimation. The audio signal analysis evaluates the performance of beamforming algorithms and of Covariance Matrix Fitting (CMF) when optimization in beamforming algorithm and alternative microphones’ configurations have been evaluated by simulations. Tests shows an accuracy in people detection of the optimized beamforming algorithm comparable to CMF method (96.5% and 96.6% respectively) with a lower computational cost. An image acquisition procedure has been then activated on the robot and the localisation of the people is performed using YOLO-v3 algorithm. Monte Carlo method applied to evaluate the propagation of uncertainty of the whole processing system presents a global accuracy of 98.2 ± 0.8%.
People detection measurement setup based on a DOA approach implemented on a sensorised social robot / Ciuffreda, I.; Battista, G.; Casaccia, S.; Revel, G. M.. - In: MEASUREMENT. SENSORS. - ISSN 2665-9174. - 25:(2023), p. 100649. [10.1016/j.measen.2022.100649]
People detection measurement setup based on a DOA approach implemented on a sensorised social robot
Battista G.;
2023-01-01
Abstract
A measurement setup for localising people in indoor environment based on a system characterized by acquisition of audio files and images implemented on a sensorised social robot is proposed. The audio signal processing for human voice identification applies the segmentation methodology for the direction of arrival (DOA) estimation. The audio signal analysis evaluates the performance of beamforming algorithms and of Covariance Matrix Fitting (CMF) when optimization in beamforming algorithm and alternative microphones’ configurations have been evaluated by simulations. Tests shows an accuracy in people detection of the optimized beamforming algorithm comparable to CMF method (96.5% and 96.6% respectively) with a lower computational cost. An image acquisition procedure has been then activated on the robot and the localisation of the people is performed using YOLO-v3 algorithm. Monte Carlo method applied to evaluate the propagation of uncertainty of the whole processing system presents a global accuracy of 98.2 ± 0.8%.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
