Accurately estimating the 6D pose of objects from single RGB images remains a pivotal challenge in computer vision and robotics. In practical industrial applications, the target object’s flexibility often complicates obtaining images with accurate pose annotations. In this work, we propose an innovative approach designed to recognize the 6D pose of industrial objects using their readily available CAD models—a valuable resource in industrial settings. Comparing regions of interest (ROIs) with rendered sparse templates, our method obtains initial rough poses of the target object. Subsequently, these estimates undergo refinement through iterative matching methods to achieve the final object pose. Experimental evaluations conducted on the T-LESS and IPPD datasets demonstrate the effectiveness of our approach. Notably, our method significantly reduces reference time without compromising accuracy when compared to prior methods. These findings underscore the efficiency of our approach, holding promise for diverse industrial applications demanding precise 6D object pose estimation.