<p>This paper presents a robust method for accurate 3D contour localization in robotic glue spraying. The method combines deep learning with 3D point cloud processing. It removes the need for precise part positioning or fixturing. As a result, the system works efficiently with arbitrarily oriented components. First, 2D image segmentation identifies the target object and creates a binary mask. This mask helps extract the 3D point cloud of target object from the scene. Keypoints predicted by a pose estimation model are used to estimate the object’s pose. This allows to calculate the initial transformation between the sample and the detected object. Next, the Iterative Closest Point (ICP) algorithm aligns the point clouds for accurate registration. The computed transformation is then used to transfer the predefined 3D contour to the target object. Experiments show that the method achieves sub-millimeter accuracy. It provides a practical and scalable solution for robotic glue spraying application on complex surfaces.</p>