Machine Vision-Based Dimensional Measurement of Industrial Parts with Circular Features

Non-contact dimensional measurement using machine vision is critical for high-precision manufacturing, yet faces challenges in accuracy under industrial constraints. This paper presents a vision measurement system built on OpenCV that automates the detection of linear and circular features in machined parts. The hardware platform integrates a 5MP CMOS industrial camera, 16mm lens, and backlight illumination. Key innovations include an optimized Zhang’s calibration method achieving distortion coefficients of [–0.12, 0.24, 0, 0, –0.08], and a multi-stage image processing pipeline combining median filtering, binary thresholding, Canny edge detection, and Hough circle/line transforms. Experimental validation using a calibrated caliper (accuracy ±0.01mm) shows measurement errors of 0.812mm for linear features (94.922mm ground truth) and 3.33–7.98% for hole diameters (10.971–11.010mm). The system achieves a mean error of 32.5μm per millimeter, demonstrating viability for industrial quality control.