Pose Estimation for Distracted Driver Detection Using Deep Convolutional Neural Networks

Distracted driver has been a major issue in today’s world with more than 1.25 million road incidents of fatality. Almost 20% of all the vehicle crashes occur due to distracted driver. We attempt to create a warning system which will make the driver attentive again. This paper focuses on a simple yet effective Convolutional Neural Network technique which can help us to detect if the driver is safely driving or is distracted which is a binary classification task. It would help in improving the safety measures of the driver and vehicle. We propose two techniques for distracted driver detection achieving state of the art results. We achieve an accuracy of 96.16% for the 10 class classification. We propose to deconstruct the problem into a binary classification problem and achieve an accuracy of 99.12% for the same. We take advantage of recent techniques of transfer learning combined with regularization techniques to achieve these results.