Estimating Traffic Density Using Sounds of Moving Vehicles

Abstract

This paper proposed a method for automatically estimating traffic density by using sounds of moving vehicles. The approach is based on the idea of recognition of the temporal variations that appear on the power signals when vehicles pass through an observation point. The local temporal variations in small periods of time are extracted by wavelet transformation and are used as an observation sequence for a hidden Markov model, which models the global temporal variations of the power signal. The passages of vehicles are detected based on the state transitions of the HMM. The occlusion problem due to the overlapping of the sounds of moving vehicles are dealt with by corresponding two set of information from a stereo microphone. Experimental results show that with some restrictions, the passages of vehicles are able to be detected from road traffic sounds in good accuracy, by the proposed method.