Research On Fatigue Driving Assistance Early Warning Based On Hadoop

With the vigorous advancement of the pace of urban modernization construction,traffic safety issues have received increasing attention from relevant regulatory authorities.Every year,so many people in the country lose their precious lives due to the fatigue of driving drivers.At the same time,the fatigue of driving drivers also causes significant economic and property losses to society and individuals.Nowadays,intelligent monitoring has become an important means to ensure the driving safety of drivers.Vehicle management companies produce a large number of driver monitoring videos every day.How to store these surveillance video data and how to extract effective information accurately and quickly from these massive surveillance videos is a problem that vehicle management companies need to solve urgently.Therefore,on the basis of these problems,this article uses the industry's popular Hadoop distributed framework to perform distributed processing on these videos,combined with the relevant information of the driver's driving behavior stored in the database,and uses information fusion algorithm to determine whether the driver is fatigued in the monitored video.On this basis,a big data visualization platform is established to manage vehicle operation and related data statistics.This article combines the Hadoop distributed processing framework and computer vision technology,uploads the surveillance video stored in the hard disk recorder to HDFS,and combines the vehicle driving information stored in the system database to determine whether the driver is fatigued.First,it mainly introduces the components of the distributed processing framework Map Reduce and the life cycle of the Map Reduce program,etc.,studies the architecture of the distributed file system HDFS and its reading and writing process,discusses the advantages and disadvantages of the HBase distributed database,and briefly introduces the message Middleware Kafka and Image Processing Toolkit Open CV.Secondly,it mainly discusses how to use Kafka to send and upload vehicle operation information to a remote server and how the remote server receives this information and saves it to the database.Surveillance video processing is to upload the video to the distributed file system HDFS through the Map Reduce module in the Hadoop framework and then cut the video,send these data to the Map function of Map Reduce,and use the information fusion-based fatigue detection algorithm to detect the fatigue driving,and save the detection results to the HBase database in the reduce function for later display.Therefore,a big data visualization platform is designed and built to monitor the running status of the entire vehicle and historical data statistics,so as to save the management cost of managers.Furthermore,this paper discusses the fatigue detection algorithm based on information fusion.Because the result of image information detection may be inaccurate,so the detection information of the driver's driving behavior is fused in the detection process.The determination of fatigue features mainly selects the driver 's eye movement characteristics and the steering wheel angle information of the driving vehicle to use the support vector machine algorithm to fuse these features.Among them,the eye movement characteristics are mainly calculated by processing the video frames through Open CV PERCLOS eigenvalues,blinking frequency and closed-eye speed,the steering wheel information of the driving vehicle is mainly collected through the CAN bus device installed on the vehicle and uploaded to the server for collection.The experiment proves that the fusion model of support vector machine is used in fatigue driving detection The accuracy rate is 92.5%.Finally,the entire fatigue driving assistance early warning system is built,the specific implementation of each module of the visualization platform is introduced,the relevant functional interface of the visualization platform is displayed,and the function of processing video for fatigue detection in single-node,pseudo-distributed and multi-node systems is also tested.Experiments show that using Hadoop clusters to process video can increase the speed by more than twice.