A study on outdoor localization method based on deep learning using model-based received power estimation data of low power wireless tag

Abstract

We are developing a method to acquire position information of a cow outdoors using Received Signal Strength Indicator (RSSI) of Bluetooth Low Energy (BLE). As existing research, there is a localization method using fingerprint database as learning data in deep learning. However, that method has the problem that it costs to create a database by measurement in a vast outdoor environment. Therefore, we considered to build a part of the fingerprint database using virtual space modeling received power measurement environment in a pasture. Experimental results showed that an average distance error to GPS data is about 6 m by training DNN using the database and additionally training DNN using actual GPS data.

Monitoring of Cow Location in a Barn by an Open-Source, Low-Cost, Low-Energy Bluetooth Tag System

“…A self-designed Bluetooth-based system was used by [8] with 4.2 m of accuracy. Outdoor animal location was monitored by a system combining GPS tags for global positioning and BLE tags for the identification and approximate localization of animals equipped with BLE tags [9], and a BLE-based system [10].…”

“…Information about the environment structure was used to eliminate impossible locations [8,19,28]. Instead of describing the environment features, a preliminary mapping (fingerprints) of RSS from several points in the environment was performed [10,18,29], though this method required time-consuming mapping, and its accuracy can be restricted by the resolution of the sampled points.…”

A study on outdoor localization method by recurrent deep learning based on time series of received signal strength from low power wireless tag

“…Once the neural network has been trained on fingerprint data relation-ships in the pasture, it can estimate positions much faster than the matching approach. For example, in [6], a virtual space is developed to mimic a pasture for grazing cattle. Then, a DNN is trained using fingerprint data generated from the virtual space and then fine-tuned by those taken from the real environment.…”

“…As a result, it is shown that this deep learning approach achieves localization error of about 6 m on average. See [6] for more details. Resultant trails by [6], however, appears somewhat awkward like zigzag walking as shown later.…”

“…See [6] for more details. Resultant trails by [6], however, appears somewhat awkward like zigzag walking as shown later. This will cause an unexpected effects on the analysis of the interactions among grazing cattle.…”