Research On Driving Behavior Recognition Algorithm Based On Deep Learning

Driver image behavior recognition based on deep learning is to extract relevant features from driver images through deep learning network for driving behavior recognition.Compared with the traditional video-based behavior recognition algorithm,the driver image recognition algorithm based on deep learning is more suitable for the actual scene.Firstly,this paper evaluates the performance of several mainstream convolutional neural networks through experiments.The experimental results show that the mainstream convolutional neural network model does not perform well in solving the problem of driving behavior recognition.Therefore,based on the characteristics of driving behavior recognition,this paper has made some improvements to the VGG-19 and RCNN model frameworks and designed three new driving behavior recognition models.The main contents of the paper are as follows.(1)Evaluation of four mainstream convolutional neural network models.Convolutional neural networks have excellent performance in the field of object classification,and driving behavior recognition is also a classification problem.Therefore,this paper first studies the four most representative convolutional neural networks in recent years,and evaluates their performance by driving behavior recognition accuracy and recognition time.(2)Research on Multi-region Feature Learning Model Based on RCNN.Experiments show that the mainstream convolutional neural network model does not perform well in solving the problem of driving behavior recognition.The main reason is that although the driving behavior recognition problem is a classification problem,the difference in characteristics between different driving behaviors is small,which is different from the traditional object classification problem.In order to further improve the accuracy of driving behavior recognition,this paper draws on the solution of image fine-grain classification problem,and improves the candidate region generation algorithm and convolutional neural network module of RCNN model framework,and designs a multi-region feature learning model based on RCNN.The model combines the driver's head,the steering wheel,and the feature information of the global image to identify driving behavior.The experimental results show that compared with the mainstream convolutional neural network model,the model can significantly improve the accuracy of driving behavior recognition.(3)Research on Driving Behavior Recognition Algorithm Based on Key Points of Human Body.Although the multi-region feature learning model based on RCNN has significantly improved the accuracy of driving behavior recognition,this method sometimes has the problem of local area feature information loss,which has a certain impact on the accuracy of driving behavior recognition.Aiming at this problem,this paper uses the idea of RCNN multi-region feature learning model,combines the human key point location model with the improved VGG-19 model,and designs a driving behavior recognition model based on human key points.The model combines multiple human key feature features with global features for driving behavior recognition.Experiments show that compared with the RCNN-based multi-region feature learning model,the model further improves the accuracy of driving behavior recognition.(4)Research on Driving Behavior Recognition Algorithm Based on Fine Classification of Key Points of Human Body.The driving behavior recognition model based on the key points of the human body is still low in recognition accuracy on some driving behaviors,and the learning ability of the model needs to be further improved.This paper draws on the idea of introducing the middle-level features into the model in the related literature,and designs a driving behavior recognition model based on the classification of key points of human body.Based on the driving behavior recognition model based on the characteristics of human key points,the model introduces the action category features of key points.According to different feature fusion methods,the model has three structural forms.The experimental results show that the introduction of the action category features of key points can effectively enhance the learning ability of the model and significantly improve the accuracy of driving behavior recognition.