Performance of convolutional neural network and recurrent neural network for anticipation of driver's conduct

Abstract

With the advancements in Internet of Things (IoT), we could efficiently improve our daily life activities like health care, monitoring, transportation, smart homes etc. Artificial Intelligence along with Machine learning has played a very supportive role to analyze various situations and take decisions accordingly. Maneuver anticipation supplements existing Advance Driver Assistance Systems (ADAS) by anticipating mishaps and giving drivers more opportunity to respond to road circumstances proactively. The capacity to sort the driver conduct is extremely beneficial for advance driver assistance system (ADAS). Deep learning solutions would further be an endeavor of for driving conduct recognition. A technique for distinguishing driver's conduct is imperative to help operative mode transition between the driver and independent vehicles. We propose a novel approach of dissecting driver's conduct by using Convolutional Neural Network (CNN), Recurrent Neural Network(RNN) and a combination of Convolutional Neural Network with Long-Short Term Memory (LSTM) that would give better results in less response time. We are likewise proposing to concentrate high level features and interpretable features depicting complex driving examples by trying CNN, RNN and then CNN with LSTM. We could improve the system accuracy to 95% by combining CNN with LSTM.