Research On Distracted Driving Behavior Recognition Method Based On Deep Learning

In recent years,with the development of China’s economy and the improvement of people’s living standards,the number of motor vehicles in China has been increasing year by year,which brings more and more road safety problems.Among them,driver distracted driving is one of the main causes of road traffic accidents,and this behaviour is usually caused by inattentiveness.In recent years,the detection of distracted drivers has been a hot topic of research.Many scholars at home and abroad have conducted a lot of research on distracted driving,specifically on aspects such as classification of distracted behaviour,detection methods and early warring models.However,there are still many problems in real-world applications:for example,traditional algorithms do not have high recognition accuracy in real-world environments;many deep learning-based detection methods do not fully consider the impact of actual hardware equipment on detection speed,resulting in poor real-time performance.To address these issues,this paper designs a deep learning-based algorithm for recognising distracted driver behaviour and transplants the algorithm to an edge computing hardware Raspberry Pi device.The main work of this paper is in the following two areas:(1)A deep learning-based distracted driving behaviour recognition algorithm is designed and implemented.Firstly,the ResNet50 network is selected as the backbone network through comparison experiments,followed by the introduction of a multi-scale fusion mechanism.This mechanism obtains feature maps of different sizes by hierarchical extraction and performs a maximum pooling operation.Next,the RFB sensory field module is added to extract a wider range of contextual information to cope with the complex environment.Finally,the SE attention mechanism module is added to the network to focus precisely on distracted driving action features and fuse them with other features to improve classification accuracy.In this paper,the method achieves the following results in the "Districted-Driver-Detection" and"ZJUT-Districted-Driver-Detection" driving behaviour datasets respectively(2)Design and development of the method.(2)A Raspberry Pi-based distracted driving detection system was designed and developed.The system uses the "Raspberry Pi+Remote Server" and "Raspberry Pi+Neural Computation Stick" architectures.In the networked state,the "Raspberry Pi+Remote Server" architecture is used,where the Raspberry Pi captures images of drivers and transmits them to the remote server via http protocol for classification,and the detection results are returned to the Raspberry Pi.In the offline state using the "Raspberry Pi+Neural Computation Stick" architecture,the Raspberry Pi captures the images and identifies the driver’s distracting actions locally.The final detection results are tallied and presented as a wave chart and bar chart.The system achieved 14.92 frames per second for the "Districted-Driver-Detection" dataset and 12.92 frames per second for the "ZJUT-Districted-Driver-Detection" dataset in the networked state." dataset reached 12.65 fps.Offline,4.06 fps was achieved for the "Districted-Driver-Detection" dataset and 3.06 fps for the "ZJUT-Districted-Driver-Detection" dataset.The system achieved 3.89 fps for the "ZJUT-Districted-Driver-Detection" dataset.The experimental results show that the deep learning-based distracted driving behaviour recognition algorithm proposed in this paper achieves a high recognition rate;the developed Raspberry Pi-based distracted driving detection system has obtained more satisfactory results through several experiments,and has good practicality and real-time recognition of the driver’s distracted driving behaviour.