A Non-contact Vehicle Weight-in-motion Method Based on Large Vision Model and Machine Learning Methods
To address the limitations in generalization capability and environmental adaptability of existing vehicle weighing systems based on computer vision,this paper proposes a non-contact weight-in-motion method based on large vision model and machine learning.Initially,an innovative edge detection algorithm using vision large model was designed for the precise acquisition of tire deformation parameters.Subsequently,a universal recognition model comprising a complete tire character database was developed based on Mask-RCNN(Mask-Region-based Convolutional Neural Networks),capable of accurately labeling and extracting features from the tire sidewalls.Furthermore,a machine learning model for predicting the contact force between vehicle tires and the road surface was constructed using LightGBM(Light Gradient Boosting Machine),and its accuracy,feasibility,and effectiveness were verified.Notably,through testing on SUV vehicles and heavy-duty trucks,this method has been demonstrated to have a maximum error of less than 5％compared to traditional weight-in-motion systems,showcasing high precision and superior performance.This study introduces a vehicle weighing technology that does not require the installation of sensors,offering a user-friendly and cost-effective solution.The potential applications of this technology are extensive,particularly within the realms of toll stations and highways,and other traffic engineering sectors.
NETL
NSTL
万方数据
