Host: Eastern Asia Society for Transportation Studies
Pages 372
The purpose of this paper is to develop traffic accident detection algorithm for improving performance of the automatic traffic accident system that are operated in Korea from 2004. The current automatic traffic accident recorder has problems of detection rates and correct detection rate decrease according to sensitivities by environmental influences and the time elapsed. In order to solve these problems, we developed the improved algorithms by utilizing of the object recognition, learning intelligence system and fluid approximation algorithm. The field test result for 2 weeks amounts to 28 movie file detection and the total number of traffic accident was 6 in actual detection areas during evaluation period. We conclude that the traffic detection rate (DR) is 100%, correct detection rate (CDR) is 21.4% and this means the performance of current traffic accident detection system (DR : 66.7%, CDR : 2.03%) is improved.