Distracted Driver Detection Using Deep Learning

Abstract

Distracted drivers are the main cause of road accidents in today's world. With the technological revolution, there are hundred of gadgets and devices that can gain the attention of a driving person and may cause serious damage to vehicle. This problem is more likely to increase as more gadgets and wireless accessories will make their way into vehicles, hence increasing the chances of driver's distraction causing serious hazards. In some unfortunate cases, distracted driver can prove fatal for themselves and for others on the road. This research is focused to address the most widely occurred problem of every country in the world i.e. the distracted behaviours of the drivers. Considering the importance and severity of the problem, we have proposed a unique yet efficient system that can detect the attentiveness of the driver. Our system not only detects the distraction of a driver but it also categorizes it into I 0 different classes. Using the dataset provided by StateFarm insurance company, we have devised a combination of two very popular deep learning algorithms i.e. CNN and LSTM. Inspired by the remarkable results of features extraction by CNN on images dataset, we assigned CNN to focus on learning the features and used LSTM to classify our data on the bases of those extracted features. By putting these two models in place our system outperforms all the models preceding our work and proposed solution was able to achieve accuracy of 99 .61 %. The results were achieved by suggesting an optimised model without distorting or enhancing the selected dataset to improve our results.