Study Of Bridge Weigh-in-motion Method Based On Dual Camera Vision And Axle Position

Monitoring traffic loads is vital for ensuring bridge safety and overload controlling.Bridge weigh-in-motion(BWIM)is currently one of the most potential new methods for moving vehicle load identification.It has the advantages of good long-term stability and convenience in installation and maintenance.However,there are still bottlenecks with this method in applicability and accuracy,especially in those scenarios that vehicles running at a non-constant speed,or multiple vehicle presence.To overcome these issues,this thesis proposes a bridge weigh-in-motion method based on dual camera vision and axle position,named V-BWIM.It can effectively solve the difficulties of vehicle positioning on the bridge deck,calibrating bridge influence lines accurately,and estimating axle weight and gross weight under the circumstances of non-constant speed and multi-vehicle presence.Therefore,it provides strong technical support for overloaded vehicle load monito ring and bridge safety operation in our country.The main work conducted in this thesis includes four parts as follows:(1)Theoretical study of V-BWIM method was conducted.A technical route for real-time positioning of vehicle axle positions based on binocular vision is established,and the pixel coordinates of vehicles and wheels in video frames are converted to actual spatial positions by a moving target detection algorithm and binocular ranging and coordinate conversion techniques.The influence line calibration and vehicle weight recognition algorithm based on axle position is developed,which overcomes the limitation of the traditional algorithm relying on the assumption of constant vehicle speed and can eliminate the negative impact of the speed change of the calibrati ng vehicle on the calibration results,and also solves the multi-vehicle weight recognition problem.(2)A comparative study on the performance of moving target detection algorithms was carried out.The experimental results show that: the traditional algorithm has outstanding advantages in lightweight model,but poor detection accuracy;Mask R-CNN has better detection accuracy but slower computation speed;YOLOv5s has the best comprehensive performance,and the image processing speed is significantly improved under GPU acceleration,which can take into account the requirements of accurate detection,real-time computation and lightweight model.In the actual application process,the most suitable target detection algorithm should be matched to the V-BWIM system according to the field conditions.(3)To evaluate the performance of the V-BWIM,relevant laboratory experiments were conducted.The results demonstrate that the V-BWIM method can provide accurate information on the wheelbase and axle position of the vehicles crossing the bridge,and can accurately calibrate the bridge influence line when the speed of the calibrating vehicle changes.The traditional BWIM method,on the other hand,will have significant errors in the influence line calibration results when calibrating vehicle speed changes,and the larger the vehicle speed changes,the larger the errors will be.In terms of vehicle weight identification,the V-BWIM method based on accurate axle positions outperforms the traditional BWIM method based on axle detectors in both single-vehicle and multi-vehicle situations.(4)In order to further verify the practical performance of the V-BWIM,a real bridge test study was carried out.The research shows that the V-BWIM method can accurately identify the axle positions and can complete the real bridge influence line calibration well,and the calibration effect is not affected by the vehicle type,speed and weight of the calibrating vehicle.In all the non-constant speed and multi-vehicle conditions of the two vehicles(three-axle and four-axle),the relative error of axle weight identification of V-BWIM method is controlled below 8%,and the relative error of gross weight identification is less than 5%.This indicates that the V-BWIM method proposed in this thesis can be a competitive alternative for future traffic load monitoring.