Article Preview
TopIntroduction
Demands for pervasive and mobile computing systems are increasing exponentially, in particular in smartphones. Indeed, such demands have made it practical to provide Location-Based Services (LBSs) like IPS and navigation (Wu et al., 2012). To provide a high level of accuracy is challenging due to the complexity of indoor localization data. Different algorithms have been proposed by researchers, and different estimations of accuracy have been obtained using RSS, Bluetooth, radio-frequency IDs, ultra-wideband, or ultrasonic technologies (Wang et., 2015). Due to its availability and lower cost, the utilization of Wi-Fi received signal strength (RSS) has become the focus of research in recent years.
RSS-based IPS techniques have been classified into two major techniques: Lateration and fingerprinting-based techniques. In general, the lateration method often suffers from inaccurate estimation. It was reported in (Wang et., 2015) that within the typical office of a length of 200 ft. and a width of 80 ft., the average localization error is about 24.73 ft. This may be due to the mulitpath propagation model (PM) due to the non-line of sight challenge and the sensitivity to errors in even one AP coordinates estimate; Thus, instead of dealing with the propagation model problem to estimate the location of the object, a radio map was prebuilt to use in a fingerprinting-based localization (Talvitie et al., 2015) scheme.
In general, the fingerprinting-based localization consists of two phases: an offline and online phase. The offline phase collects the RSS readings with their time sampling and their location to generate a prior fingerprint at the fingerprinting database, which contains the reference points (RPs). The number of RPs has a direct impact performance of fingerprinting-based methods. The online phase estimates the actual location by comparing the RSS value of a mobile device with the predefined fingerprints by using an IPS. One of the simplest ways to estimate the mobile user’s location is the k nearest neighbor algorithm (kNN), which estimates the localization by computing the k nearest neighbors that have the smallest Euclidean distance between the two phases (Bahl and Padmanabhan, 2000). Such an algorithm has low accuracy but is easy to implement. In this work, we propose:
- •
A Probabilistic Neural Network-Jensen-Bregman Divergence (PNN-JBD) for a WLAN-based method. We perform the matching stage using probability kernels as a regression scheme;
- •
A procedure with high characterization distribution to be used. RSS value was taken in four different orientations (45°, 135°, 225°, and 315°) to prevent body-blocking effects, with ten scans with a delay of 10 seconds to reduce the effect of signal variation;
- •
PNN-JBD results outperforms the results of PNN and kNN with respect to accuracy and the average error distance, which indicates that the proposed combining scheme is more effective in sensitive environments of WLAN-based positioning systems.
TopThere are two main types of IPS methods: lateration and fingerprinting. The lateration method uses log-distance to describe the relationship between mobile devices to multiple access points (APs) (Chen et al., 2006). However, due to the complexity of the indoor localization environment, the lateration-based localization is not accurate and the distance estimations are often with large errors. such as if one AP coordinate has been taken inaccurately this can drop the performance dramatically. So, to ensure the localization performance, the filter techniques have been used; for example, in (Wang et al., 2012) the particle filter was proposed, while in (Choa & Kim, 2010) the Kalman filter was used and in (Dhital et al.,2012) the Bayesian filter was used to ensure the performance by restricting the localization error through trace movements.