Analysis of Driving Skills based on Deep Learning using Stacked Autoencoders

Takuya Kagawa, Naiwala P. Chandrasiri


Due to the advancement of automobile technology and increasing consumers demands, it is expected that automatic driving vehicles and manual driving vehicles will coexist in future automobile society. There are a number of people who are interested in driving and, they may think that the automatic driving vehicles are unnecessary. However, if the vehicle is operated manually, there is a possibility for driving skills of a driver to fluctuate due to drowsiness and fatigue and that may lead to accidents. In such a situation, it is important for vehicle to monitor the driver's driving conditions and provide with a driving support system or automatic driving options. In this research, we propose a method to classify driving skills of an individual driver with high precision based on deep learning (stacked autoencoders). In the experiments, driver’s driving skills were classified by combining sensor signals of curve driving data acquired from a driving simulator. As a result, a maximum driving skill recognition rate of 98.1% was achieved. In addition, the recognition rate was improved compared to the previous researches.

Full Text: PDF


  • There are currently no refbacks.