Research of the performance improvement in traffic accident detecting system within the cross road

The purpose of this paper is to develop incident detection algorithm for improving performance of the automated traffic accident recorder in operation in Korea from 2004. The current automated traffic accident recorder has problems in increase of detection rate(DR) and correct detection rate(CDR) that are sensitive to environmental influences as time elapsed. In order to solve these problems, we developed the improved algorithms by utilizing of the object recognition, grant of weight and fluid approximation algorithm, the result of field test for 2 weeks amounted to 28 detections of the movie files, total numbers of real traffic accidents was 6 cases in the detection area during the evaluation period. As the test was 100% of detection rate(DR), 21.4% of correct detection rate(CDR), we learned the system was improved the performance than existing system(DR 66.7% CDR 2.03%).