A retail application based on indoor location with grid estimations

Indoor positioning systems have many technological varieties and application scenarios. Through development of a LBS service for targeted retail scenario we have adopted Wi-Fi signal strength fingerprinting considering cost, resolution and scaling-up factors. To tackle the accuracy problem of returned location coordinates and make the LBS service more impactful, we have looked into past studies of probabilistic smoothing techniques such as Bayesian modeling, clustering and regression neural network. The robustness of these techniques was found sensitive to signal attenuation due to the complexity of indoor setup and dynamics of on-site crowds. We have therefore come out with a grid estimation method for the collected coordinates on top of Bayesian smoothing, which was proved to be simple but effective for the proposed retail scenario.