Research And Implementation Of Dangerous Driving Behavior Identification System Based On Deep Learning

With the popularization of vehicles,road traffic accidents have become the focus of attention,and most of them are mainly caused by dangerous driving behavior.Dangerous driving behavior mainly consists of bad driving behavior and uneven road conditions.Therefore,online monitoring of these two aspects of the situation to improve road traffic safety has a very important significance.Because smartphones have the advantages of low cost and high portability,this paper will use smartphone sensors to conduct real-time monitoring.Based on the above background,this paper proposes a dangerous driving behavior recognition system based on deep learning.The specific work is presented as follows:(1)this paper proposes an abnormal driving behavior recognition algorithm based on TCN and soft thresholding,which can solve the noise and real-time problems of current mobile phone sensors.The algorithm takes advantage of TCN’s fast training and prediction speed and the filtering property of soft threshold function.A soft attention mechanism is introduced to selectively use sensor information related to driving behavior.Based on this model,a road condition detection algorithm based on multi-features is designed,which can obtain road condition information while detecting driving behavior.The multi-feature refers to the driver’s behavior characteristic and the road attribute characteristic.(2)This paper constructs two platforms of a dangerous driving behavior system.The Web management platform is a unified management platform for enterprise managers to manage users,data and other information and monitor data in real time.The Android APP is a platform that allows drivers to give real-time warning of bad driving behaviors and bad road conditions and carry out personalized route planning.Finally,the proposed algorithm of abnormal driving behavior recognition and road condition detection is tested in four public datasets.The results show that the algorithm of abnormal driving behavior recognition has higher computational efficiency and the accuracy is improved by 2.24%compared with the best SOTA model.The accuracy of both algorithms in online tests was above 97%.The functional effects of the Web management platform and Android APP in this paper also meet the expected effects.