Determine Traffic Accidents Based on Changes in Driving Patterns

Abstract

A visual surveillance system in detecting accident occurrences is crucial to minimize the risk of fatalities caused by delays in handling accident victims. This research aims to identify accidents by observing changes in driving patterns through CCTV video recordings. The research method utilizes machine learning techniques to learn the features of the direction and speed of vehicles under normal traffic conditions and during accidents. Farneback optical flow obtains each tracked vehicle object’s velocity and direction values. The classification algorithm then studies these data to build a model for determining accident occurrences. Based on the testing results, the model achieved an accuracy of 86%, precision of 92%, and recall of 85% using the Random Forest algorithm. Additionally, the classification performance evaluation using the ROC curve yielded an AUC value of 86% for Random Forest. Thus, the speed and direction feature of vehicles provide valuable information about driving patterns, serving as parameters in determining accident occurrences.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint