Driver distraction causes one of the major problems in road safety and accidents. According to the World Health Organization (WHO), over 285,000 estimated accidents happened as a result of distracted drivers per year. To address such a fatal issue and considering the future of Intelligent Transport System, we have proposed a novel ensemble learning approach based on deep learning techniques for detecting a distracted driver. In the proposed approach, we have fine-tuned the Faster-RCNN for detecting the objects involved in distracting the driver during driving and achieved 97.7% validation accuracy. Moreover, to make the prediction strong and reduced the false positive, pose points of the driver have also extracted. By using those pose points, we make sure that we detect only those objects which are directly associated with the driver’s distraction. The interactive association of various objects with the driver has calculated using the intersection over the union between the detected object and the current posture features of the driver. Our proposed ensemble learning technique has achieved over 92.2% accuracy which is far better than previously proposed models. The proposed method is not only time-efficient, robust, but cost-efficient as well. Such a model not only can ensure road safety as well as help Governments to save resources being spent on monetary losses.
A Novel Ensemble Learning Approach of Deep Learning Techniques to Monitor Distracted Driver Behaviour in Real Time
2021-04-06
2566327 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch