posted on 2009-03-01, 00:00authored byHsiuping Lin, Ying Zhang, Martin Griss, Ilya Landa
Many context-aware mobile applications require a reasonably accurate and
stable estimate of a user’s location. While the Global Positioning System (GPS)
works quite well world-wide outside of buildings and urban canyons, locating an
indoor user in a real-world environment is much more problematic. Several
different approaches and technologies have been explored, some involving
specialized sensors and appliances, and others using increasingly ubiquitous Wi-
Fi and Bluetooth radios. In this project, we want to leverage existing Wi-Fi access
points (AP) and seek efficient approaches to gain usefully high room-level
accuracy of the indoor location prediction of a mobile user. The Redpin algorithm,
in particular, matches the Wi-Fi signal received with the signals in the training
data and uses the position of the closest training data as the user's current
location. However, in a congested Wi-Fi environment where many APs exist, the
standard Redpin algorithm can become confused because of the unstable radio
signals received from too many APs. In this paper, we propose several enhanced
indoor-locationing algorithms for the congested Wi-Fi environment. Different
statistical learning algorithms are compared and empirical results show that:
using more neighbors gives better results than using the 1-best neighbor;
weighting APs with the correlation between the AP visibility and the location is
better than the equally weighted AP combination, and automatic filtering noisy
APs increases the overall detection accuracy. Our experiments in a university
building show that our enhanced indoor locationing algorithms significantly
outperform the-state-of-the-art Redpin algorithm. In addition, this paper also
reports our findings on how the size of the training data, the physical size of the
room and the number of APs affect the accuracy of indoor locationing.