Recognizing object localization using acoustic markers with active acoustic sensing

We propose a low-cost sensing system that recognizes an object’s location on a surface using active acoustic sensing. Our proposed system uses a thin speaker attached to an object as a marker and estimates the marker’s location from the sound source. Localization is achieved through machine learning (random forest) based on the property that high-frequency components of sound decrease more than low-frequency components with distance. We additionally implemented a system to simulate the condition where multiple objects are placed simultaneously and to estimate the frequency response of those objects from training data where only a single object is placed. Performance tests show that our system localizes a single object with a mean absolute error of 0.41 cm in a 20 cm square area on a wooden deck and also localizes the placement of four objects with an accuracy of 1.83 cm while saving 83.3% of the effort needed to collect the training data.