Abstract The integration of artificial intelligence into industrial processes has significantly improved workflow efficiency, particularly for in-line machine vision inspection systems. These systems depend on precise data acquisition and advanced processing for reliable measurements. A key challenge in the acquisition phase is determining the camera-object spatial relationship, typically achieved through extrinsic camera calibration. However, in single-camera setups, this calibration often relies on external markers, which can compromise accuracy and introduce measurement errors. This paper presents a practical approach for estimating objects dimensions based on camera-object distance as a complement to intrinsic camera calibration, hence eliminating the need for external markers. The reliability of the proposed approach is validated through a statistical analysis. An additional challenge addressed in the present study resides in the processing phase which suffers from the absence of standardized approaches, which affects feature detection reliability. To deal with this shortcoming, a selection of widely employed classical and deep learning-based processing techniques is evaluated to assess their impact on feature extraction. The findings provide valuable insights for developing a structured qualification methodology in camera calibration and processing selection, improving inspection efficiency in machine vision applications.
Taatali et al. (Fri,) studied this question.