RF-Based Vehicle Detection and Speed Estimation

Abstract

Developing a robust and reliable vehicle detection and speed estimation system that alerts drivers about driving conditions and helps them avoid joining traffic jams is an important problem that has attracted lots of attention recently. In this paper, we introduce a novel RF-based vehicle motion detection and speed estimation system (ReVISE). Our system leverages the fact that the presence of objects in an RF environment affects the received signal strength and hence, can be used to detect and identify different characteristics of the objects in an area of interest. Our long-term vision for ReVISE is to leverage common wireless networks, such as WiFi or cellular, to detect the density of traffic and estimate the car speed based on the mobile devices carried by users. This gives us an edge over the current techniques for traffic estimation as we do not require any specialized hardware and the cellular signal strength information is available from all cell phones, providing large-scale ubiquitous traffic estimation. We present the design and analysis of ReVISE including its vehicle detection and speed estimation modules. The detection module can differentiate between an empty street, stationary cars, and moving cars based on a multi-class SVM approach that uses features from the RF signal strength. We also present two novel speed estimation techniques based on statistical and curve fitting approaches. Evaluation of ReVISE in a real testbed shows that the proposed techniques can detect vehicle motion with an accuracy of 100% and estimate the vehicle speed with an accuracy of 90% in typical streets. This highlights the feasibility and promise of using RF for vehicle detection and speed estimation.