Roadside Estimation of a Vehicle’s Center of Gravity Height Based on an Improved Single-Stage Detection Algorithm and Regression Prediction Technology

In this paper, a vehicle center of gravity (CG) height estimation method based on improved You Only Look Once algorithm and AdaBoost regression is proposed by relying on roadside sensors. The CG height of a vehicle plays an important role in the performance evaluation indexes of vehicle handling, ride comfort, and safety. Therefore, the accurate measurement of the CG height is helpful for improving vehicle control and ensuring driving safety. Different from traditional vehicle modeling, in this study, a data-driven roadside estimation method based on machine vision is utilized for the estimation of the vehicle CG height, thereby providing an economic and convenient concept. In terms of vehicle type identification, the improved YOLO algorithm is characterized by flexibility and speed while ensuring high accuracy. This ensures that it can be quickly deployed on roadside equipment and meet real-time requirements. Moreover, through the correlation analysis, parameters including the vehicle’s front track width, rear track width, roof height, and weight, which can be collected by roadside sensors, are selected as input parameters for the regression. An ensemble learning model based on AdaBoost regression realizes the final estimation of the CG height. The test results obtained on the NHTSA experimental data show that the average accuracy of vehicle type identification reaches 97.59% and the regression estimation accuracy of the vehicle CG height reaches 94.64%.