Research On Driving Identification In A Turning Based On Mixed Deep Learning

Advanced driver assistance systems are a hot topic recently.The system uses sensors installed on the body and inside to collect the external environment during driving,data during driving,and driver status to predict driving conditions.The core of the advanced driver assistance system is based on the sensor’s perception of the environment inside and outside the vehicle.With the help of deep learning,it assists the driver to drive safely and avoid risks,helping the driver to develop good driving habits.In the short turn process,many driving details are included.During the short corner driving,many driving details such as deceleration timing,turning speed,turning range,acceleration timing,etc.are included,which is enough to reflect a driver’s Behavioral habits.The driver’s driving behavior in the corner is analyzed through a deep learning network to model and identify driving behavior habits.This paper proposes a curve driving behavior recognition model suitable for experimental data.The model is a mixture of convolutional neural networks and two-way LSTM deep learning.The experimental results show that the average accuracy of the model is 81.9%.The average accuracy of LSTM is only 77.4%.(1)By defining the process of cornering driving,the data required for the experiment is accurately taken out during the pretreatment stage.The multidimensional scaling method is used to reduce the dimensionality in a variety of dimensionality reduction methods,and the Spearman correlation coefficient is used as the distance measurement method of experimental data to achieve feature extraction.Finally,better dimensionality reduction results are obtained.More features are retained in the case of less information loss.(2)It is proposed to use Bi-LSTM to solve the problems that occur in the traditional CNN and LSTM for timing signal processing.Compared to other deep learning architectures,CNN has more convolutional layers to maintain features,and a pooling layer to reduce excessive parameters.Gradient disappearance or gradient explosion problems that are difficult to solve with traditional RNNs are solved in Bi-LSTM.At the same time,LSTM is unidirectional.Bi-LSTM can make full use of the sequential logic or the coherent logic of the information to eliminate the disadvantage that the information after the current moment can not be trained.A two-way cyclic neural network combined with convolutional neural networks is established and used to identify the behavioral characteristics of curve drivers.This model preserves the excellent ability of the convolutional neural network to extract local features,and the ability of the two-way neural network to retain information for a longer period of time,improving the accuracy of driver recognition.