HIGHWAY TRAFFIC CONGESTION CLASSIFICATION USING HOLISTIC PROPERTIES

摘要

This work proposes a holistic method for highway trafficrnvideo classification based on vehicle crowd properties. Thernmethod classifies the traffic congestion into three classes:rnlight, medium and heavy. This is done by usage of averagerncrowd density and crowd speed. Firstly, the crowd densityrnis estimated by background subtraction and the crowdrnspeed is performed by pyramidal Kanade-Lucas-Tomasirn(KLT) tracker algorithm. The features classification withrnneural networks show 94.50% of accuracy on experimentalrnresults from 254 highway traffic videos of UCSD data set.