A high available and strong generalization capability data-driven-based wireless fingerprint positioning model

This is the authors' abstract. We don't add key points for this paper.

The demand for indoor location-based services is far greater than supply, and many novel indoor localization technologies are emerging. Wi-Fi fingerprint-based method has drawn much attention by virtue of its user-friendliness and no extra device deployment. However, the traditional methods based on signal similarity or received-signal-strength-based (RSS-based) machine learning face their bottlenecks, such as the performance bottleneck or the generalization capability. Thus, a binary-classification model based on sample differences is proposed, which divides the original fingerprint data into positive and negative samples. Then, new relative features are proposed to replace the traditional media-access-control(MAC)-RSS features. Finally, the classification models are trained using decision tree (DT), support vector machine (SVM), and random forest (RF) algorithms, and several sets of controlled experiments are performed to evaluate the proposed method. It is noticeable that the proposed method has better performance in floor detection accuracy and point matching accuracy, and the mean error can reach 2.9m (DT), 3.09m (SVM), and 2.96 m(RF). Besides, the proposed method shows a very excellent model generalization capability, and the model trained by one building's data can still work well in another building. Those advantages have enormous potential in many large indoor localization scenes, like mall, theme park, airports, museums and exhibition halls.