LSTM-based Head-on Collision Warning System with a Decentralized Radio Sensing Approach

This paper studies the performance of an automatic head-on vehicle collision warning system based on a decentralized sensing approach using radio frequency (RF) signals. To identify a vehicle moving toward another, a communication system is used in reception mode, and a continuous wave (CW) RF signal transmitted by a third vehicle driving behind as a probe signal. The gathered signal was classified using a long short-term memory neural network. A data set consisting of CW RF signals was collected in a series of experiments in a highway scenario. The signals were processed to find Doppler signatures of approaching vehicles. This information is used to detect events of interest. Our results demonstrate the system’s feasibility, obtaining a precision of 98.6% in a multi-class classification assessment.