Could or could not of Grid-Loc: grid BLE structure for indoor localisation system using machine learning

Indoor localisation and its various applications have received significant attention in recent years. The state-of-the-art systems include a large number of complex hardware structures and algorithms making the system not suitable for practical applications. In this paper, we integrate a localisation system that consists of device development, model deployment, data collection and localisation algorithm to explore the localisation accuracy in a special static indoor environment (i.e. a meeting room or a parking lot). Compared with previous studies, the significance of our work is to find out a more convenient and practical way to deploy devices with a simple algorithm (e.g. machine learning algorithm) in such a scenario. Besides, it is meaningful to explore the technology of indoor localisation based on the application scenario. We propose a Grid-Loc system that presents a grid structure of Bluetooth low-energy devices to collect data assisting localisation. The system is easy to deploy for reducing the signal attenuation caused by the objects’ occlusion. Meanwhile, the system applies an algorithm that combines adaptive boosting with a support vector machine algorithm to support the system. In our deployed localisation scenario, we also compare localisation performances for several algorithms; the result shows the Grid-Loc system achieves the accuracy of 91.2%, computing time within 3 s in real time and a low cost. The system is also robust and scalable under the same indoor environments.