Augmented Reality (AR)-based programming approaches hold great promise for addressing the challenges of flexible automation by facilitating fast and intuitive programming processes. Pose estimation of novel objects enhances the program-ming experience by bridging the real and virtual environments. However, a prerequisite for pose estimation is to perform a 2D segmentation to determine the region of interest (ROI). In this work, we present an AR-based approach that enables point-and-click ROI detection through human interaction. Our proof of concept investigates how the achievable accuracy varies with the quality of the user input. The results show that the accuracy of the ROI estimation has a minimal impact on the overall accuracy. Existing limitations can be addressed by other approaches presented.