High Impact Factor : 4.396 icon | Submit Manuscript Online icon |

Detecting Distracted Driver


Nishant Srivastava , IMS Engineering College; Roopak Singh, IMS Engineering College; Mandeep Singh, IMS Engineering College; Rishabh Gupta, IMS Engineering College


CNN, SVM, Keras, Theano, Tenserflow


Drivers are supposed to be focusing on driving by law. However, it is very common to see drivers doing something else while driving: texting, drinking, operating the radio, talking on the phone and etc. This distracted behaviors easily cause crash incidents. According to the report from National Center for Statistics and Analysis, each day there are over 8 people killed and 1,161 injured in crashes due to a distracted driver in US, which translates to 423,765 people injured and 2920 people killed each year. To alarm the distracted driver and better insure their clients, State Farm Insurance hopes to design an alarm system that can detect the distracted behavior of car drivers by using a dashboard camera. As a result, they held a online Kaggle competition to encourage Kaggler to build a robust computer vision system. Specifically, in this task we were provided with an image dataset which consists of 10 classes of driver behaviors. For each test image, we were required to output the probability that the image belong to each of the ten classes. Two algorithms have been tried here and compared for the performance: Support Vector Machine (SVM) and Convolution Neural Network (CNN). To supplement the training set, pseudo-label semi-supervised technique is used. We also implemented a recently-developed CNN structure called VGG-GAP for visualizing what the neural network is looking for in the task, so as to better analyze the learned pattern and search for improvements. This task is very meaningful for improving the drivers safety and can be easily applied to other applications such as triggering autonomous driving and etc.

Other Details

Paper ID: IJSRDV7I21138
Published in: Volume : 7, Issue : 2
Publication Date: 01/05/2019
Page(s): 1321-1322

Article Preview

Download Article