Indoor cell-level localization based on RSSI classification

Abstract

The task of estimating the location of a mobile transceiver using the Received Signal Strength Indication (RSSI) values of radio transmissions is an inference problem. Contextual information, i.e., if the target is in a specific region, is sufficient for most applications. Therefore, instead of estimating position coordinates, we take a slightly different approach and look at localization as a classification problem. We perform a comparison between the K-Nearest Neighbor (KNN), the Support Vector Machine (SVM) and the Simple Gaussian Classifier (SGC), three classifiers proposed previously under different contexts. Using experimental results, we demonstrate that the SGC achieves a competitive performance despite its simplicity. Furthermore, we consider the extension of the SGC to a Hidden Markov Model (HMM) and demonstrate the performance gains. The derivative of the HMM filter allows us to do online parameter tracking, realizing an adaptive scheme. To our knowledge, this adaptive scheme has not been used for the SGC before. Considering the advantages of the SGC, we advocate the SGC as a competitive solution for estimating contextual location information.

Direction Estimation in 3D Outdoor Air–Air Wireless Channels through Machine Learning

“…For each method, we utilize the same features and dataset as in the proposed method for training and testing. In addition, we also evaluate the performance of the SVM method provided in [35] for direction estimation in the A2A link, where the authors in [35] utilized the received signal strength of the access point node as a feature in the SVM classifier for indoor localization. The direction estimation accuracy of each approach is presented in Table 3, along with the associated mathematical representation of prediction complexity and the number of operations required for predictions, where T is the depth of the tree in the decision tree method, A n and A s are the total number of layers and average number of neurons per layer in the ANN method, respectively, and D is the number of training samples in the KNN method.…”

“…The comparisons show the superiority of the proposed SVM method in terms of accuracy. The SVM model from [35] provides the lowest accuracy of all because it only uses the received signal strength of a single communication path between the transmitter and receiver as a feature to estimate the direction. The KNN method provides comparable accuracy to the proposed method but at the cost of high complexity, which scales with the number of training samples.…”

RSSI-Based Machine Learning with Pre- and Post-Processing for Cell-Localization in IWSNs

“…The main task in such use cases is to classify specific environments or regions like the inside or outside of a room and determine whether a sensor node belongs to such a confined region. Authors in [7], [8] have studied the so-called cell-level-based localization with RSSI values using supervised machine learning methods. A major challenge with these methods is the limited amount of training and validation data.…”

Frictionless indoor location using BLE

“…After that, several classification algorithms like SVM, random forest, neural networks, etc, can be tested to ensure the accuracy of the algorithm. In [28] a simple Gaussian classifier was used and the result was improved by using Hidden Markov Models, as the position of a person (cellphone) in the grid depends on its last position. The result of this process is not an x,y position of the person, but the cell where it's located.…”