Design And Implement Of Distraction Driving Detection System

Driving is a complex task that requires a high concentration of attention.It includes information sensing,decision making and execution.In a transportation system consisting of a driver,a vehicle and a road environment,the driver is the weakest link in the system.Distracted driving is caused by inattention caused by tasks that are not related to driving tasks(such as adjusting the radio,making a call,texting,etc.).In the distracted state,the driver's safety awareness decreases,and the shift of attention reduces the driver's ability to perceive,judge,and make decisions about the surrounding environment,affecting the safe handling of the vehicle,and easily causing traffic conflicts.To this end,we designed and implemented a distracting driving real-time monitoring system based on the Movidius neural computing stick,which can quickly identify the driver's distraction state,and find that the driver is in dangerous driving behavior for a long time(2-3 seconds).The status will alert the driver.At the same time,if necessary,the monitoring results can be transmitted to the cloud platform,so that the traffic control center personnel can remotely grasp the driver status.The Movidius Neural Computation Bar is a USB-based deep learning inference tool and a stand-alone artificial intelligence accelerator that provides dedicated deep neural network computational acceleration for embedded devices,enabling edge devices to deploy highly accurate neural network models.Firstly,by modifying the existing GoogLeNet network model,it can be trained to identify dangerous driving behaviors.Then,the dangerous driving behavior monitoring model is combined with our proposed information fusion-based distracting driving detection method to deploy to Movidius.In the end,in order to strengthen the management capabilities of the traffic control center personnel,we upload the test results to the OneNet cloud platform and display them in real time,so that the management center personnel can view the driver status in real time.The experimental results show that the Raspberry Pi equipped with the Movidius neural computing stick has a processing time of about 90 milliseconds for the driver's picture,and the processing speed of the whole system is 5-8 frames per second,which basically realizes the task of real-time monitoring of distracting driving.