Terrain Classification For Ground Robots Based On Acoustic Features

Abstract

The motivation of our work is to detect different<br> terrain types traversed by a robot based on acoustic data from the<br> robot-terrain interaction. Different acoustic features and classifiers<br> were investigated, such as Mel-frequency cepstral coefficient and<br> Gamma-tone frequency cepstral coefficient for the feature extraction,<br> and Gaussian mixture model and Feed forward neural network for the<br> classification. We analyze the system’s performance by comparing<br> our proposed techniques with some other features surveyed from<br> distinct related works. We achieve precision and recall values between<br> 87% and 100% per class, and an average accuracy at 95.2%. We also<br> study the effect of varying audio chunk size in the application phase<br> of the models and find only a mild impact on performance.