| The two-wheeled vehicle elevator prohibition system in this paper is based on the background of the rapid increase in the number of electric vehicles and the frequent spontaneous combustion and explosion accidents of electric vehicles,in order to prevent some people from driving two-wheeled vehicles into office buildings and highrise residential buildings through elevators for charging and storage,Avoid blocking the building safety channel and eliminate the potential risk of spontaneous combustion and explosion of electric vehicles,and protect the safety of people and property in buildings and residences,this paper presents a two-wheeled vehicle elevator entry prohibition system based on deep learning to detect inside the elevator electric cars,bicycles and electric bicycle,prevent it into the elevator.In order to avoid the negligence of human monitoring and prevent the safety risks brought by the twowheeled electric vehicle entering the elevator from the source,it is of high research significance and practical value to design a feasible and reliable two-wheeled vehicle elevator entry prohibition system.The main contents of this paper are as follows:(1)In this paper,a deep learning-based entry prohibition system for two-wheel vehicle elevator is proposed.By extracting the monitoring picture in the elevator,the median filter is adopted to reduce the noise in the video image,and the deep learning target detection technology is used to detect the no-entry target.Once the no-entry target is identified,the system will trigger the alarm and emergency stop device to remind passengers,it also prevents the elevator from running and prevents two-wheeled vehicles from entering the elevator.(2)Data sets are collected,annotated,and enhanced.5316 data pictures were shot and collected,LabelImg was used to annotate the data pictures,and 15948 related pictures were extended through data enhancement to complete the production of the data set,and determine the performance evaluation index of the algorithm evaluation.(3)In order to solve the problems of YOLO V4 model being too large to be well deployed in embedded devices and slow recognition speed,the backbone network of YOLO V4 algorithm was replaced by MobileNetV2 in this paper,and the PANet structure was optimized by using deep detachably convolution blocks,thus completing the lightweight improvement of the network.It can meet the application scenarios of the no-entry system,accelerate the speed of target detection and recognition,and reduce the parameters of the lightweight model by 83.26%.In order to compensate for the loss of precision after lightweight,this paper improves the accuracy of positioning frame by improving the position regression function.Focal loss function was introduced to solve the problem of sample imbalance.By adding CBAM attention mechanism,more effective feature information can be obtained,and mAP before and after improvement is increased by 2.98%.(4)The improved YOLO V4 algorithm is applied to the entry prohibition system of the two-wheeled vehicle elevator to complete the test and analysis of the no-entry system.When the no-entry target is detected,the alarm and emergency stop device can operate correctly.The mAP of the two-wheeled vehicle elevator entry prohibition system model in this paper can reach 95.3%,the inference speed is 0.0268 seconds per piece,and the model size is 39 MB.Compared with the original YOLO V4 algorithm,although the model mAP is reduced by 1.69%,the inference speed is increased by 40.2%,the model size is reduced by 84%,which can meet the detection and deployment requirements of the two-wheeled vehicle elevator prohibition system in this paper.Through test,the facades system proves that the two vehicles elevator forbid system can achieve the desired effect. |
