Distracted Behavior Detection of Commercial Vehicle Drivers Based on the MobileViT-CA Model
Due to the particularity of their occupation,commercial vehicle drivers are prone to distracted driving behavior during driving,resulting in major traffic accidents.In order to improve the detection accuracy and generalization of distracted driving behavior of commercial vehicle drivers,we proposed a driver distraction behavior detection method based on the improved MobileViT network.Based on the natural driving real vehicle tests,we constructed a dataset of distracted behavior of commercial vehicles,including safe driving,using phone,drinking,hair or makeup and talking to copilot.Then,the attention mechanism was introduced into the lightweight MobileViT network.And the optimal classification model MobileViT-CA was designed by selecting effective network backbone MobileViT,attention module CA,and network embedding position.The research results show that the MobileViT-CA classification model proposed in this paper can effectively improve the performance of the classification network,and the accuracy of the distraction behavior dataset of commercial vehicle drivers and the State Farm dataset under normal lighting conditions reaches 96.57％and 99.89％,respectively.Meanwhile the model has the advantages of small size,high detection accuracy,high reliability and generalization ability.
NETL
NSTL
万方数据
维普
