Research On Recognition Algorithm Of Dangerous Movement Based On Deep Learning

In recent years,with the rapid increase in the number of cars in China,the frequency of traffic accidents has also increased.Relevant researchers found that most of the causes of traffic accidents are related to drivers,and nearly half of them are caused by dangerous driving.Therefore,it is an urgent problem to identify and avoid drivers' dangerous driving behaviors.Intelligent onboard devices can be used to monitor the driver's dangerous driving behavior in real time,and then warn the driver or force him to reduce the speed.The reduction of traffic accident rate can protect people's life and property from loss,and also can effectively alleviate the problem of road congestion during rush hours.The main research contents of this paper are as follows:First,kinect device was used to collect the data sets of 5 driving states of 15 drivers.At the same time,the data sets of 5 driving states of 26 drivers in the open source State Farm was mixed to produce the data set required for the experiment.Then the driver is divided into training set and test set according to the ratio of 4:1 to ensure the objectivity and effectiveness of the algorithm results.After that,aiming at the problem that the driver image collected by vehicle equipment contains a lot of redundant information,which may lead to the low accuracy of dangerous driving behavior identification,this paper proposes an algorithm to locate the driver's local information to help the convolutional neural network to identify dangerous driving behavior.According to the local location information of the driver,the interference of redundant features can be reduced.At the same time,convolutional networks can focus on the key characteristics that determine driving behavior.Finally,aiming at the problem that the driver image collected by vehicle equipment contains a large amount of redundant information,which may lead to low accuracy of dangerous driving behavior identification,an optimization method to reduce redundant information is further proposed,namely,a method for dangerous behavior identification based on codecdecoding structure semantic segmentation model and BAM attention mechanism.In conclusion,this paper conducts experiments and analyses from the perspective of reducing redundant feature information of images and screening effective features of dangerous driving behaviors,which not only improves the accuracy of vehicular intelligent devices in identifying dangerous driving behaviors,but also improves the convergence speed of convolutional neural network in identifying dangerous driving behaviors.The experimental results show that the improved method can accurately detect the driver's dangerous driving behavior.