The major advantage of products made from composite materials, such as carbon fiber reinforced polymers (CFRP), is given by their superior weight-specific, mechanical properties such as strength and stiffness. These properties can be weakened by defects induced in the manufacturing process. In dry fiber filament winding, online detection and analysis of the processed fiber bundle geometry is a key factor for the quality assurance of the final part. In this work, the instrumentation and data evaluation for determining the surface geometry of fiber bundles by means of a light sectioning sensor was examined. Profiles of glass fibers were measured continuously in a specifically developed inspection test rig. By application of an interactive polynomial fitting algorithm, data segmentation of object and base line was robustly achieved on varying background conditions. In addition, unwanted defects as well as lateral movement of the fiber bundles were reliably detected. The information revealed by the proposed algorithm provides the basis for robust online monitoring of fiber bundle geometry in highly automated composite manufacturing processes.