Research On Driver Smoking Detection Algorithm Based On Deep Learning

In the context of rapid social and economic development,people have higher pursuits and expectations for the quality of life.As a convenient,comfortable and safe means of transportation,cars have become the main choice for public travel.Cars provide convenience for people’s travel,but with the continuous increase of car ownership,traffic safety issues have become increasingly prominent,and various traffic accidents have occurred frequently,causing serious losses to people’s lives and property.Cars provide convenience for people’s travel,but with the continuous increase of car ownership,traffic safety issues have become increasingly prominent,and various traffic accidents have occurred frequently,causing serious losses to people’s lives and property.Among them,dangerous driving behaviors such as smoking by drivers are one of the important causes of traffic accidents.This behavior not only affects the safety of the driver itself,but also threatens the safety of other road users.Therefore,it is extremely important to supervise the dangerous driving behaviors of drivers such as smoking.As deep learning and computer vision related technologies advance rapidly,detection algorithms such as target recognition and behavior recognition are applied in various fields,and the algorithms for smoking behavior detection are also evolving.However,the existing smoking detection algorithms generally have low detection accuracy,slow detection speed and missed detection,and their performance in practical application is not satisfactory.In view of the above problem,this paper proposes an improved YCAM smoking detection algorithm based on YOLOX,and applies the algorithm to the intelligent forklift management system of Yinzhou District to assist the forklift supervision department to supervise the smoking behavior of forklift drivers.The main work of this paper is as follows:(1)Construct a smoking behavior detection dataset.The smart forklift all-in-one machine is installed on the forklifts of several companies in Yinzhou District,Ningbo City,and the driving behavior of the forklift driver is filmed through the network camera,and then the captured video is captured into pictures frame by frame using Open CV.Then,related operations such as classification and sorting,data enhancement,and data labeling are performed on the pictures to obtain a smoking behavior data set.(2)Improved object detection algorithm.To address the problem of low detection accuracy of small targets in the conventional target detection algorithm,this paper designs an improved YCAM smoking detection model based on the YOLOX algorithm.This paper first adds the convolutional attention mechanism CBAM to the backbone network to make the detection network allocate more attention to where it is needed,and then changes the original positioning loss function Io U-Loss to CIo U-Loss to solve the problem of gradient disappearance,while using Focal-Loss to regress the confidence of the detection target to alleviate the problem of unbalanced positive and negative samples.Finally,BCE-Loss is used to judge the category of objects in feature points.The accuracy rate of the improved YCAM smoking detection algorithm is verified by experiments,the accuracy rate is 96.53%,the m AP value is 98.69%,the recall rate is 97.35%,and the speed is 15.73 ms.The overall performance is better than the YOLOX algorithm,which meets the needs of practical applications.(3)Design and implement a forklift driver smoking detection system.When the driver is smoking while driving the vehicle,the system will give an alarm in time to remind the driver,and save the smoking video and pictures to the background.Assist the forklift supervision department in the supervision of forklift drivers to achieve the purpose of reducing the driver’s dangerous driving and protecting the safety of the driver’s life and property.