On Simultaneous Recognition and Locating of Contextual Objects with Unmodulated Acoustic Signals Using Machine Learning

This paper examines the problem of simultaneously recognizing and locating contextual objects via using unmodulated acoustic signals from the working contextual objects recorded by the COTS smartphones of users. We exploit the frequency and power features of these signals to build a Mel-frequency cepstral coefficients (MFCC) dataset for contextual objects, and constructs a classifier for contextual object recognition by using BiLSTM and a regression model for object-to-device distance computation by using LightGBM, which is then used for object locating with the help of user’s trace. Extensive experiments show that the proposed approach achieves high recognition accuracy and locating accuracy, even when there are ambient noises or the sampling time span of accoustic signals varies.