Fast Hierarchical Template Matching Strategy for Real-Time Pose Estimation of Texture-Less Objects

This paper proposes a fast template matching strategy for real-time pose estimation of texture-less objects in a single camera image. The key novelty is the hierarchical searching strategy through a template pyramid. Firstly, a model database whose templates are stored in a hierarchical pyramid need to be generated offline. The online hierarchical searching through the template pyramid is computed to collect all candidate templates which pass the similarity measure criterion. All of the template candidates and their child neighbor templates are tracked down to the next lower level of pyramid. This hierarchical tracking process is repeated until all template candidates have been tracked down to the lowest level of pyramid. The experimental result shows that the runtime of template matching procedure in the state-of-the-art LINE2D approach [1] after applying our fast hierarchical searching strategy is 44 times faster than the original LINE2D searching method while the database contains 15120 templates.