Real-time Single Detector Vehicle Classification

Abstract

The estimation of speed from a single vehicle detector has been a popular area of research largely due to the potential for reduced maintenance and installation costs associated with the more conventional method of using two closely spaced detectors with a known offset distance. Speed cannot be directly measured using a single detector but flow and occupancy measurements (the percentage of time a detector is occupied in any given sample) can be recorded and used to estimate the speed and subsequently, using the vehicle presence time, the length of vehicles.The approximation of space occupancy from the time occupancy provided by a single short detector requires that vehicle speeds in each sampling interval are constant and subsequently research on this subject is concentrated on freeway applications where speeds can generally be assumed constant during each sampling interval. However, in urban situations and particularly in the vicinity of traffic signal controlled junctions it cannot be assumed that this is the case due to the rapid variation in vehicle speed.Centralised on-line adaptive traffic signal optimisation strategies use data received from detectors placed on approaches upstream of junctions to feed information into an underlying traffic model. Despite the advancement in communication technology 4Hz and 10Hz protocols are still used widely in Urban Traffic Control (UTC) and consequently this research explores the accuracy of speed estimation and basic vehicle length classification that can be achieved using data sampled at these rates.For this application data must be available within a time window of a few seconds to be useful in the optimisation process. This work investigates what level of accuracy can be obtained by estimating vehicle speed and length immediately following a vehicle leaving the detection zone using a process of speed matching to provide useful data to an on-line traffic model.