Automatic Stop Line Violations Detection using Histogram of Oriented Gradients and Support Vector Machine

Abstract

This research presents an automatic system for the detection of stop line violations. Stop line violation is defined as a condition in which a vehicle does not appropriately stop behind a stop line. This research uses the Histogram of Oriented Gradients (HOG) as feature extractor and Support Vector Machine (SVM) as classifier. We tested our system using a self-made dataset, containing images taken from a traffic CCTV camera. This dataset consists of three sub-datasets for different conditions: photos taken in the morning, afternoon, and evening. Our experiments show that the proposed system achieves accuracy rates of approximately 94.2%, 98.1%, and 95.9%, respectively, for the morning, afternoon, and evening sub-datasets (using HOG cell size of 16x16, linear kernel SVM and k-fold validation with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$k$</tex> =10). Therefore, the average accuracy rate is 96.07%. The system is also shown not to be severely affected when the photos are taken during a rainy condition, yielding an average accuracy of 95.3%. Finally, when the system is tested using new data (i.e., those not involved in the k-fold validation experiments) an average accuracy of 98.7% can be achieved. These results show that the proposed system performs well and that the combination of HOG and SVM shows good potential for application in automatic detection of stop line violations.