Traditional marker-based photogrammetry systems often require the attachment and removal of a sticker for each measured point, involving labor-intensive manual steps. This paper presents an innovative approach that utilizes raised, cross-shaped markers, referred to as ‘molded markers’, directly embedded into composite pieces. In this study, these markers, commonly employed in other industrial processes, serve as fiducial markers for accurate photogrammetry. A two-stage detection algorithm is developed to accurately identify their centers: initial approximate detection by a Faster R-CNN model, followed by accurate localization using a classical cross center detection algorithm. This study investigates the pertinence of using polarimetric images to guarantee the highest detection rate and accuracy even in adverse lighting conditions. Experimental results demonstrate the viability of using these markers in accurate photogrammetry systems, achieving a median accuracy of 0.170 (interquartile range (IQR) 0.069 to 0.368) mm/m while enhancing automation and system usability.

Incorporating propagation of quantified uncertainties into the design process of close-range optical-marker-based photogrammetry solutions would help in developing more accurate and precise systems able to keep up with competing metrological solutions in emerging Intelligent-Manufacturing environments. This article presents a novel analytical model for image-coordinate measurement precision of circular markers as a function of sensor noise and detected ellipse shape and grey-value intensity. The model can predict the image coordinate measurement standard deviations required for uncertainty propagation of estimated 3D coordinates more reliably than published techniques. Synthetic images with known lens distortion, sensor noise and focusing blur are used to characterise a selection of subpixel circular marker measurement methods with particular emphasis on uncertainty propagation during eccentricity error compensation. For the first time for the existing literature, the eccentricity-error correction disambiguation uncertainty is analysed, i.e., which of the two existing candidates for projected centre must be taken. At the same time, the range of ellipse shapes where eccentricity-error correction becomes ineffective in presence of other uncertainty sources is characterised.