Research On Drowsy Driving Warning System For Mobile Terminal Based On Multi Index Information

With the rapid development of China’s economy,the number of private cars in our country has increased by more than 10%each year.The problem of driving safety has also become a stubborn disease of modern transportation.Accounting for up to 20%of the total number of traffic and showing an increasing trend year by year,Fatigue driving has become an important factor affecting safe driving that needs to be addressed.Studies have shown that 90%of the dangers of fatigue driving can be effectively avoided by warning one second in advance,so the fatigue driving warning system proposed in this thesis is of great significance for traffic safety.This thesis combines traditional machine learning algorithms and deep learning algorithms to build a terminal-oriented drowsiness warning system.The entire system can be divided into three major modules:driver’s facial information extraction module,drowsiness-related state index calculation module and drowsiness warning module.The first module contains a face detection algorithm,a key-point detection algorithm and a head-pose estimate algorithm.In this thesis,the multi-task convolutional cascade network is used to detect the driver’s face.By replacing the fully connected layer in the output network and reducing the pyramid scale,the storage space is reduced by 3 times and calculation speed is increased by 10 times while maintaining relative accuracy.The key-point detection method uses constraints local model.According to the characteristics of this algorithm,this thesis improves the algorithm speed by 3.5 times from scale optimization and parallel processing,and reduces the model storage by about 8 times.Then,the Euler angle of the driver’s head is solved according to the key-point information.The second module involves the calculation of eye amplitude,mouth amplitude calculation,and Euler angle calculation of head posture.Set thresholds for changes in the normal state,to determine whether the eyes are open and closed,whether to yawn,whether to look left and right,and whether there is a drowsy nodding.The third module combines the status information obtained by the above modules to count the corresponding four statuses in time.In order to ensure real-time early warning and at the same time offset errors caused by inaccurate positioning of key points,a one-second time window is used to monitor the drowsiness state to trigger corresponding alarm information.Experiments show that the drowsiness warning system constructed in this thesis has an average accuracy rate of 91.79%in the four warning messages of closed-eye detection,yawn detection,nodding detection,and distraction detection,which meets the accuracy requirements of the drowsiness detection system.Moreover,the algorithm of this thesis has a processing speed of 17 frames per second on a low-cost hardware platform(Qualcomm MSM8909),which satisfies the real-time requirements of the drowsiness warning system under resource constraints.In summary,the drowsiness warning system constructed in this thesis has high practical application value.