Segmentation and detection of object points on non-textured surfaces are one of the basic image processing processes in some photogrammetric applications. In this paper, we approach this problem in two steps. 1). A local thresholding process is applied which detects local maximum contrast. From this we get a primary segmented binary image. 2). Several contour tracing and thinning processes followed which in the meantime remove noises and the traced circles whose perimeters lie outside an interactively given range. Thus almost all the object points are detected. The whole procedure can be controlled interactively based on certain a priori knowledge and application requirements. This approach has demonstrated good performance with application to images taken from the photogrammetric test field at the institute.
H. B. Zhou,
"Object points detection in a photogrammetric test field", Proc. SPIE 1395, Close-Range Photogrammetry Meets Machine Vision, 13954D (1 August 1990); doi: 10.1117/12.2294400; https://doi.org/10.1117/12.2294400