ABSTRACT
In this paper, an asset localization method is proposed, which is based on the Direction-Of-Arrival (DOA) estimation method with a single beacon and a Convolutional Neural Network (CNN). It is used for smartphone users to locate an asset which a Bluetooth Low Energy (BLE) beacon is attached to. The DOA corresponds to the target direction of the strongest Received Signal Strength (RSS) so it can meet the requirement of finding the asset’s orientation. The proposed target localization method only depends on a single beacon, while a conventional DOA scheme needs an antenna array. While the user is looking for an asset, the orientation of the smartphone and the RSS will be changed over time. Thus, the CNN model identifies the orientation of asset relative to the user, according to the continuous time series information acquired by the smartphone. At each sampling time point, a set of the temporal data consists of the RSS of the BLE beacon and sensor data of the Inertial Measurement Unit (IMU). The CNN model can extract features and higher-level information, which is from the spatial relationship among different types of sensors and temporal relationship among the same sensor. Through the training process, the CNN model can identify the orientation of the asset. In this research, TensorFlow is adopted to construct our CNN model. Two experiments were conducted to evaluate the localization angle error: open space and indoor environment with shadowing and reflection. The experimental result showed that the localization angle error is less than 25 degrees. It indicates that the proposed method is able to fulfill the needs of asset localization.
