Counting and locating people in outdoor environments: a comparative experimental study using WiFi-based passive methods

WiFi-based passive methods are becoming a common tool to count, estimate, and/or locate people. One area of applicability is the development of intelligent control system for traffic management in urban areas, so that these systems are able to take into account not only vehicles' behaviors but also pedestrians', as important actors in the road scenario. In this work, we present the performance evaluation in terms of accuracy of a WiFi-based passive method used to identify pedestrians, classify them as moving pedestrians or static pedestrians, and for the latter, to locate them in a traffic intersection. The proposed algorithm is implemented in a low-cost development board and tested through several experiments in a real outdoor scenario. Our proposal is compared with several classic Machine Learning (ML) algorithms, specifically with Binary Logistic Regression, Support Vector Classification, Gaussian Naive Bayes, Random Forest, and k-Nearest Neighbors. Results show that despite the simplicity of our method, the outcomes are similar or better than most of the ML techniques, without the expected complexity or computational requirements that the latter required.