Factors affecting the accuracy of a computer vision-based vehicle weight measurement system

Vehicle weight is an important element of traffic data as it affects both traffic safety and service life of infrastructures. Traditional techniques for vehicle weight measurement, such as static weighing and weigh-inmotion, are costly to maintain and have limited service life. Thus, the non-contact vehicle weight measurement has emerged as a promising alternative. This study proposes a computer vision (CV)-based method for vehicle weight measurement and the factors affecting its accuracy are analyzed. The process of CV-based weight measurement involves the capturing of tire (tyre) images, processing them to extract tire edges, calculating vertical tire deformations, determining actual tire pressures, and using the tire mechanics model to calculate the weight of each tire and the gross weight of the vehicle. Parametric studies are conducted to study the effects of factors such as distance between camera and tire, camera parameters, illumination, background, and vehicle speed on measurement accuracy. The results indicate that high accuracy is achieved under optimal conditions, such as short distance, appropriate camera parameters, normal illumination, large feature difference between tires and background, and low speed. In application, the identification error of gross vehicle weight is within 15% at 95% confidence interval, which is comparable to a weigh-in-motion system.