Research About Indoor Positioning Based On WLAN

Internet companies at home and abroad have paid attention to Location-Based Service(LBS). Essentially, Location-Based Service is a kind of value-added services. An important part of Location-Based Service is locating consumers' position. Due to buildings' blocking and interference, what we can utilize in outdoor environment performs poorly. This means we need to find a special method to handle indoor positioning.In these years, smart phone became popular. And hardware configuration of smart phone is better and better, which makes smart phone owns a capability to implement some complex computation tasks.Wireless Local Area Network(WLAN) based on 802.11 series protocols has covered places such as dwellings, plot, office, super market, airport and exhibition. Indoor positioning technique based on WLAN makes sufficient use of the wireless access points(APs) off the shelf. And smart phone have ability to implement positioning algorithms. So it is no need to install extra devices which is made for positioning. Comparing with other techniques used for indoor positioning, indoor positioning based on WLAN have advantages.This article introduces some basic algorithms in indoor positioning, and discusses some classical indoor position systems: Active Badge, RADAR, Mote Track, Cricket, Horus and my COEX. Then explain some algorithms that are used in this article which contains k-Nearest-Neighbors, weighted k-Nearest-Neighbors, the Kalman Filter and it variations. Method this article use to implement positioning algorithms is location fingerprint. First, making efforts to implement multi-Gaussian model, this article finds and analyses some problems, and explain the reason that this method does not fit the data collected in this experiment. And then switching to k Nearest Neighbors method and weighted k Nearest Neighbors method to estimate position. Using RSS average to build radio map. After that, using Kalman Filter to track and correct user's position. In this article, weights of wk NN are standardized to eliminate large scale error. And this article uses a linear model in Kalman Filter, and its variable of speed is estimated through several former positioning results.Comparing different results caused by differentk values in k NN method, different k values and different kernel functions which contains 8 common kernel functions in wk NN, this article finds the optimal parameter values and kernel functions. Using these optimal parameter values, the mean error before applying Kalman Filter has been limited within 2m. Integrally, mean value of wk NN's positioning error and its variance are smaller than k NN's.Finally, applying Kalman Filter which is found no use to significantlyincrease accuracy while it can smooth out positioning error and decrease error average and variance.