Human Localization Using Multi-Source Heterogeneous Data in Indoor Environments

Abstract

A human localization system using multi-source heterogeneous data in indoor environments is proposed in this paper. The system can be easily constructed with already deployed Wi-Fi and camera infrastructures and is able to make use of received signal strength samples, surveillance images, and room map information to achieve a comparable performance. In a corridor scenario, we optimize propagation model (PM) parameters with crowdsourcing data from only several locations and establish a data table of optimized parameters for trilateration localization. These crowdsourcing data are also used to correct trilateration localization results, through which localization performance can be greatly improved. In a room scenario, we locate a human object with a panoramic camera and room map. We first detect the human object on the observed image and search a pixel location that represents the object’s location best. Then, the pixel location on the image is mapped to the room map using an artificial neural network. By this method, localization accuracy of sub-meter level can be obtained. We perform the proposed system in our experimental environment, and the experimental results show that our localization system not only requires no extensive time and labor cost, but also outperforms fingerprinting and PM localization systems.