Research And Implementation Of Electric Bicycle Violation Automatic Detection System

As the development of cities and towns continues to accelerate,traffic also becomes increasingly congested.Therefore,"small,fast,flexible,and labor-saving" electric bicycles are gradually being used as transportation by more and more people.However,due to the increasing number of electric bicycles,followed by some problems,such as riding an electric bike without wearing safety helmets,not installing license plates,overload and so on.These violations are very easy to lead to traffic accidents.Traffic police law enforcement costs a lot of manpower.Although the phenomenon of violations will decrease after traffic polices spot check in different places,it will rebound immediately after the supervision is lax.In order to solve the above problems,it is necessary to study the automatic detection of electric bicycle violations.The "electronic police" can greatly improve the supervision efficiency of electric bicycle violations,and realize effective supervision at any time and place.The implementation of the object detection function in this paper uses the YOLOv3 algorithm,and the YOLOv3 is optimized.For the data samples in this paper,K-means++ algorithm is used to cluster the annotation boxes,the original anchor boxes are replaced by the newly obtained anchor boxes,Io U is replaced by DIo U to calculate the loss,the convolutional layer and BN layer are combined,and NMS is replaced by Soft-NMS.By comparing and analyzing the experimental results of the above optimization methods,it is found that m AP and average Io U are improved after replacing anchor boxes,the speed is increased by 6% after combining the convolutional layer and BN layer,the average Io U is increased by 2.6% after using DIo U.When the NMS was replaced with Soft-NMS,although the overall performance decreased,the AP and average Io U are improves for person class.Through the requirement analysis of the automatic detection system of electric bicycle violation,this system mainly realizes the automatic detection of the violation phenomenon of not wearing safety helmet,not installing license plate and overloading.If no license plate is detected,the electric bicycle does not have a license plate,which belongs to unlicensed driving;if the number of safety helmets worn on the head is less than the number of people on the electric bicycle,someone does not wear a safety helmet;if the number of people on the electric bicycle exceeds two,it is judged to be overloaded;finally,license plate text recognition is carried out for electric bicycles of violation.Using the object detection method based on YOLOv3 to realize the function modules of riding an electric bicycle detection,license plate detection,safety helmet detection and person detection.For license plate text recognition,firstly the license plate image is rotated using Hough transform.Then the algorithms CTPN and CRNN based on deep learning are combined to realize the text recognition of electric bicycle license plate in unconstrained scenarios.Finally,the automatic detection system of electric bicycle violation is displayed in the form of web page.