Date of Publication :5th January 2025

Abstract: Research on traffic light detection and recognition (TLR) is growing gradually every year. Additionally, Machine Learning (ML) has been extensively applied, not solely within TLR investigations but across various fields where data generalization and the automation of human behavior offer practical benefits. This study discusses many artificial intelligence and machine learning techniques for identifying and recognizing traffic lights. It adopts a strategy of initially classifying and then identifying. The strategy begins by locating the traffic light region, extracting it, conducting image processing tasks, and then delivering the processed image to the recognition method, The LISA dataset, an open-source resource utilized in this research, comprises 43,007 frames of continuous video sequences designated for both testing and training, along with 113,888 annotated traffic lights, To collect the dataset, a stereo camera was installed on the roof of a car, which was driven at various times, including both day and night, while encountering a range of lighting and weather conditions. The accuracy of the suggested algorithms was impressive, with the decision tree scoring 97%, logistic regression scoring 98%, and the support vector machine (SVM) achieving the highest score of 98.62%. These outcomes reflect the successful application of artificial intelligence and machine learning algorithms in effectively recognizing and predicting traffic light signals in different lighting and weather scenarios.

Reference :