Regulating traffic in urban cities is highly dependant on traffic lights, particularly at intersections, where crossing a red light could jeopardize many lives. For this specific reason it is crucial for self-driving cars to recognize traffic lights and abide by the rules that traffic lights help enforce. In this paper a deep neural network was trained to detect and classify traffic lights, moreover, the output of this model was given to an algorithm that would alert the driver in case the computer detects that the driver might cross the red light, this feature could be incorporated in current Advanced Driver Assistance Systems (ADAS) systems in a car. Additionally, a camera-based method was proposed to filter out the irrelevant traffic lights in a scene. To train the model, transfer learning based on the You Only Look Once version 4 (YOLOv4) algorithm was employed, and the LISA dataset - which includes images from streets in California, USA - was used for the training process, as for the testing process different images were used including ones from Cairo, Egypt. The model trained in this paper reached a mAP score of 92.16%.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    A YOLO Based Approach for Traffic Light Recognition for ADAS Systems


    Contributors:


    Publication date :

    2022-05-08


    Size :

    2808152 byte




    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English



    Yolo-Based Traffic Sign Recognition Algorithm

    Ming Li / Li Zhang / Linlin Li et al. | DOAJ | 2022

    Free access



    Traffic Sign Detection and Recognition for Hazy Images: ADAS

    Galgali, Raiee / Punagin, Sahana / Iyer, Nalini | Springer Verlag | 2021


    ADAS METHOD FOR ESTIMATING TRAFFIC DENSITY USING ADAS PROBE DATA

    KIM HYUNGJOO / SEO YOUNGHOON / LIM DONGHYUN | European Patent Office | 2022

    Free access