Abstract
To address the collaborative requirements of high precision, high efficiency, low cost, and non-contact measurement for wheel arch detection in the calibration of Advanced Driver Assistance Systems (ADAS) during vehicle production, this study proposes a monocular machine vision-based detection methodology. The hardware system incorporates an industrial camera, priced at approximately 1000 CNY, and a custom light source. The YOLOv5s model is employed for rapid localization of the wheel hub, while the MSER algorithm, in conjunction with Canny edge detection, is utilized for robust feature extraction of the wheel arch. A geometric computation model, referenced to the wheel hub, is subsequently established to quantify the wheel arch height. Experimental results indicate that, for seven vehicle models, the method achieves an average absolute error (MAE) of ≤0.25 mm, with a maximum error of ≤0.545 mm and a single measurement time of ≤3.2 s, making it suitable for a 60 JPH production line. Additionally, under lighting conditions ranging from 500 to 1500 lux and dust concentrations of ≤10 mg/m(3), the MAE fluctuation remains within ≤0.08 mm, ensuring consistent measurement accuracy. This methodology offers a cost-effective, reliable, and fully automated solution for wheel arch detection in ADAS calibration, demonstrating strong adaptability to production lines and considerable potential for industrial applications.