Pedestrian is one of the important entities for the urban city and various facilities are already provided by the government to make pedestrian walking more safe and comfortable, including the pelican crossing system. Pelican crossing is designed for urban area and it will stop the traffic (by changing the traffic light to red) if pedestrians press a specific button. The main problem of pelican crossing is that the crossing time is fixed and it not adjusted based on the condition of the pedestrian, e.g. number and walks speed of the pedestrian. In this paper, we propose an adaptive time arrangement system on pelican crossing using convolutional neural network (CNN) classifier. The system is built using two different cameras, with the first camera pointing to the pedestrian waiting area and other camera pointing to the pelican crossing. We utilize MobileNet-SSD (Single Shot Detector) CNN architecture that originally used for object detection problem. The MobileNet-SSD CNN classifier was trained using MS-COCO dataset for the first step and fine-tuned the weights on VOC dataset. We remove all VOC categories except for person class because the class will be used for pedestrian detection in the proposed system. The pedestrian crossing time is then calculated based on the detected pedestrian and some predefined pedestrian walk speed and start-up time. To test the proposed system, we have collected several videos that represented the real system environment and conducted experiments on those data. Experiments show that the system is feasible to use in the pelican crossing situation with some appropriate configuration recommendation.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Pelican Crossing Adaptive Time Arrangement using Convolutional Neural Network




    Erscheinungsdatum :

    2019-11-01


    Format / Umfang :

    2954587 byte




    Medientyp :

    Aufsatz (Konferenz)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch