Driver identity identification and driving state monitoring method based on machine learning and deep learning

Abstract

The invention relates to a driver identity identification and driving state monitoring method based on machine learning and deep learning. Motion data of an automobile are acquired through a smart phone sensor, thereby identifying motion of a vehicle driver. Fuzzy mode identification is utilized for dividing the motion sequence of the driver to a driving operation. Then according to road traffic information and an image photographing device, front obstacles and a jam condition in automobile driving are identified through computer vision technology, and furthermore different driving scenes are divided. According to a driving operation, statistics characteristics are respectively extracted. Furthermore a characteristic vector is formed as an input of a deep neural network. The identity of the driver is identified through constructing a personal driving characteristic database and training a corresponding deep neural network model. After the identity of the driver is confirmed, the driving state of the driver at each time is identified through a recursion neural network. According to the method of the invention, multiple-signal-source data are utilized; and based on the driving operation and the driving scene, a deep learning method is utilized for improving identification accuracy.

Description

technical field [0001] The invention relates to the field of vehicle driving operation based on the built-in inertial sensor of the smartphone and the identification of driver identity and driving state, in particular to a method in the field of driver identity and driving state identification based on machine learning and deep learning. Background technique [0002] The current state-of-the-art driver identification technology is based on a convolutional neural network with a fixed-size sliding time window combined with a recurrent neural network model (CNN+RNN). The disadvantage of this method is that before the features extracted by convolution are input into the training model of machine learning, the discriminative degree of the features for the identity of the driver is unknown. In addition, due to the use of a fixed-size sliding time window, it is easy for a driving operation, such as turning, to be mechanically divided into two time windows to extract features for tr...

AI Technical Summary

Problems solved by technology

But similarly, since the same driver will have different performances when performing the same action in different driving scenarios, it is unreasonable to directly use the original data to detect the driving state, and the recognition accuracy is also very low.

Images