Design And Implementation Of Drivingfatigue Testing Application Based On Deep Learning

On the basis of a large number of road traffic accidents,the incidence of traffic accidents was greatly increased by drunk driving,speeding and fatigue driving.Driving fatigue is a serious problem,which is considered to be a direct or indirect cause of road traffic accidents.In addition,the testing technology and products of driving fatigue are not well applied,so it is necessary to develop the detection technology of fatigue driving.Firstly,fatigue characteristics were extracted from images by detection technology,including facial fatigue features,human fatigue characteristics and head movement characteristics.In the process of extracting eigenvector,the improved algorithm of Canny face detection was mainly used to extract features and integrate the results into matrix.The algorithm was high accuracy and real-time performance,and it was popular used in the field of face detection algorithm.Then on the basis of DBN deep network model,the model training was trained by unsupervised learning and a greedy algorithm,DBNL model was built.According to the four characteristic vectors,driver fatigue driving was judged by image.In order to verify the accuracy of the models,the BP-ANN model which is often used in classification of judgment and Naive Bayes model were trained and compared with DBNL model,the results showed the DBNL model effectively improve the accuracy of determination.Next,aiming at how fatigue driving influence traffic safety problem,according to the combination of multiple matter model,a kind of new risk quantitative discriminant model which was combined with the Deep learning DBN algorithm and classification analysis method was put forward,the model was used to produce warnings to the driver and reduce the number of traffic accidents.Using Logic algorithm as the output layer classifier,DBNL discriminant model and general training method based on sample model are established.The recognition accuracy of the model was 93.78%,it was increased by 20.11%and 14.45%respectively compared with Naive Bayes and BP-ANN neural networks.Finally,by adjusting the core parameters of the DBNL model,the results showed that the training model was more stable and had better reliability.In summary,the DBNL model can effectively discriminate fatigue driving,it also can help to realize the occurrence of automatic identification fatigue driving and give the driver warning information after the occurrence.