Circle and arrow traffic light recognition
Stephen Stephen, Dwi Hendratmo Widyantoro · 2017
Car accident has been one of the accidents that occur the most in every part of the world and one of the main reasons is the negligence of the driver whether he is feeling sleepy or not paying attention to the traffic light. In this paper, we build a prototype that is able to tell the driver in advance about the upcoming traffic lights so he/she will be able react appropriately. The main steps for traffic light recognitions include color segmentation, feature extraction using Histogram of Oriented Gradients (HOG), training the features using Support Vector Machine (SVM), Artificial Neural Network (ANN), and Random Forest (RF), then classification. Performance was evaluated using 10-fold cross validation using videos taken during morning and night. The experiment results reveal that the best model is delivered using SVM with polynomial kernel. Its accuracy achieves 98.5% for morning data and 98.8% for night data.