Traffic light candidate elimination based on position

There have been some traffic light detection systems developed by researchers and most of them use camera to detect traffic light. Sometimes, those systems still detect many non-traffic light objects as traffic lights. This paper presents methods for eliminating traffic light candidate based on the traffic light candidate's position in video's frame by using machine learning algorithm. The traffic light candidates are represented by bounding boxes. The candidate features include the bounding box position, its dimension as well as its ratio, relative to the video's frame size. It first trains the model of traffic light position using training data and then predicts the traffic light candidate using the same classifier as employed during learning the model. Experiment results show that the system could eliminate traffic light candidate based on position with 0.42 Precision and 0.92 Recall with Random Forest algorithm.