A Deep Neural Inferencing Approach of Assistive Philippine Traffic Light Recognition: An Augmented Transfer Learning Approach

Study on the identification of traffic signals plays an important role not only for intelligent cars but also for traditional cars and their drivers. Various identification methods have been suggested over the years for traffic light identification, unfortunately not in the Philippine environments. In this study, the proponents developed a traffic light recognition system that could be used in crossroads/crosswalks in the Philippines for traffic light recognition. The researchers trained and tested the dataset using Yolo v3 and resulted in a mean average precision (mAP) of 0.88. The system has been tested by traffic lights in the Philippine environment, which has a good result in terms of traffic light detection and recognition.