Identification of People in a Camera's Field of View Using Acoustic Signal from Mobile Phone

Abstract

We propose a graph matching based approach to identify multiple people in an indoor localization system by using the infrastructure camera along with microphone and speaker in a smart phone. Given a frame captured by the camera, an attributed graph is composed by considering each person in the frame as the vertice and the distance between people derived from the locations provided by the camera as the attribute of an edge. In the same way, another attributed graph is obtained from the perspective of smart phone, which has a unique ID and enables acoustic ranging technique between the people. Since more accurate distances are from camera than from smart phone, the graphs on the camera and smart phone are respectively defined as model graph and data graph for graph matching method. For working in real time, Spectral Matching algorithm is applied on the two graphs. As well, we propose to make use of slide window in the graph matching method in order to improve the approach's robustness, which indicates it can provide reasonable identification results even with much erroneous distances. The excellent performance of the proposed approach is demonstrated by results obtained from several sets of test-driven simulation experiments.