Non-contact Fatigue Driving Detection Method Based On Millimeter Wave

In the era of global economic interconnection,people’s living standards have gradually improved and social development has accelerated.People need to travel long distances frequently for work,study,and entertainment,which has led to increasing use of transportation,resulting in frequent traffic accidents,and fatigue driving is a key factor in causing traffic accidents.In order to reduce this potential safety risk,some fatigue driving detection methods have emerged to achieve real-time monitoring and early warning of fatigue driving;however,existing methods are usually contact-based,easily interfering with the driver’s driving process,and the research methods lack some accurate fatigue identification indicators.In this study,a non-contact fatigue driving detection method based on machine learning is proposed.Using the advantages of millimeter wave with high-speed rate,high accuracy,low energy consumption and high interference resistance to achieve non-contact accurate detection of fatigue driving.The main work of this study is summarized as follows.(1)We propose a fatigue action detection method based on convolutional neural networks for the problem of high accuracy perception of fatigue actions of people in different environments.In the human target detection stage,the MTI clutter suppression algorithm is used to effectively suppress clutter and reduce environmental interference,and the CA-CFAR detection algorithm is used to determine the required detection targets on the radar path without interference from false targets,improving the target detection performance and enhancing the stability and generalizability of the system to achieve effective fatigue action detection of targets,after which the micro-Doppler of fatigue action is collected feature maps for fatigue action classification recognition.In the experimental part,a comparative analysis of the system performance was conducted on subjects at different detection ranges,different scenes and different angles of equipment placement,and the results showed that the method has a high accuracy rate for fatigue action classification,with an average detection rate of 94.13%.(2)We propose a fatigue detection method that combines fatigue actions and breathing signals,for the problem of dynamic combined perception of fine-grained and coarse-grained behaviors.In the fatigue action detection stage,the micro-Doppler feature map of the action is collected and input into an adapted convolutional neural network for recognition and classification,and the breathing detection stage is entered when the fatigue action is detected.In the breathing detection stage,the breathing signal is obtained by separating from the obtained original signal,and then the background noise is removed by band-pass filtering and moving average filtering to obtain a smooth breathing waveform,and using a pseudo-peak removal algorithm that removes the pseudo-peak,and the peak detection algorithm is used to estimate the respiration rate to achieve fatigue detection.The recognition rate of this method can reach 91.81%.Meanwhile,the system is confirmed to have good robustness through multiple groups of comparison experiments.(3)We propose a method for fatigue driving detection using the time domain and frequency domain features of breathing signals and heartbeat signals.The system first uses millimeter wave radar to collect raw data in a real vehicle environment,separates and reconstructs the breathing and heartbeat signals,and after smoothing and processing,obtains effective feature signal data,and selects seven effective features of breathing signals and heartbeat signals: average heart rate value,r MSSD,heartbeat low frequency,heartbeat high frequency,heartbeat low frequency/heartbeat high frequency,average breathing frequency and heart rate/breathing value.The feature regressivity analysis was performed to obtain the feature nodes of human fatigue,further labeling the data,and the fatigue driving detection could be performed in real time after training the random forest model,and the average recognition accuracy of fatigue driving could reach 93.21%,and the robustness of the system was further verified by subsequent comparison experiments.