This paper presents a novel method for a precise localization of the automated-fiber-placement head, without the need for a data access to the machine control. It is based on a sub-pixel accurate optical-flow-algorithm which determines information about the heads movement by means of the material flow in sequences of IR images. Using local curvatures in the temperature field of the IR images, feature matrices are created which can locally be compared to the features of successive images. Thus, the translation between images become visible. This enables the possibility to perform an accurate (16.8μm) and self-sufficient process monitoring that additionally is capable of capturing the motion and position information of the AFP system and can be linked to existing algorithms for defect detection and classification. © 2021, The Author(s).