Efficient localization using different mean offset models in Gaussian processes

Indoor positioning using wireless signal strength has become an area of highly active research. Many papers prior to this one have demonstrated how Gaussian processes can be used to generate a likelihood model for signal strength measurements. One advantage of Gaussian processes is the ability to efficiently calibrate devices by using SLAM technique. However, Gaussian process is, by default, a zero mean process, which doesn't reflect the true nature of signal propagation. In many works, algorithms are modified to use a constant, non-zero mean offset. There is also a modification using a simple mean offset model where signal strength decreases linearly with the distance from the access point. In this paper, a log-distance radio propagation model as a mean offset model for Gaussian processes is proposed. This model was chosen since many works have demonstrated it's a good correlation to experimental results. Amongst the three models described, the log-distance model provides the highest accuracy. Also a comparison was made with the k-nearest neighbor method and probabilistic histogram approach, showing the superiority of methods using Gaussian processes. All algorithms were optimized which made it possible to perform all calculations on a mobile phone in real time.