Signals of Opportunity for Car Density Estimation with Limited Training Data

Passive sensing leverages existing signals from illuminators of opportunity to perform target localization, detection, and tracking without the need for additional infrastructure. In this paper, we explore the use of WiFi signals of opportunity for estimating car density in parking lots. The aim is to develop an efficient and cost-effective parking occupancy estimation system to alleviate traffic congestion caused by drivers searching for parking spaces. To achieve this, experimental measurements were conducted in a real outdoor environment to collect reflected WiFi signals from targets (cars). These collected signals were then used for car density estimation using a combination of semi-supervised learning convolutional neural network (CNN) and weighted-centroid interpolation techniques, only requiring small size datasets and limited measurements. The proposed method overcomes the limitations of existing data-driven estimators by reducing the reliance on large labeled datasets and computational complexity associated with traditional supervised learning methods. In addition, it provides a cost-effective alternative to the traditional systems that rely on a large number of sensors. Simulations are performed to evaluate the performance of the estimation, and the results demonstrate that our scheme can effectively estimate car densities in the parking lot with reasonable estimation errors.