Rapid and Efficient Facial Landmark Identification by Light and High Resolution Network using Artificial Intelligence

The ability to detect facial landmarks is in high demand because of its usefulness in monitoring driver health in real time. Heat map-based approaches are well-known for their exact accuracy in landmark position prediction, tiny, and excellent-resolution networks may accomplish quick estimations. While tiny high-resolution networks (THRN) have been successful in connecting feature maps of varying resolutions, the fusion block’s costly computation remains an issue. Furthermore, THRN does not make use of the powerful output module seen in high-resolution networks. We propose an entirely novel design called THRN to address these issues. Using multi-resolution feature maps and an innovative fusion block, THRN Plus improves in two ways. First, it reduces the processing timing of its output module. On two different face landmark datasets, we ran extensive trials and showed that THRN greatly improved baseline techniques while keeping the latest in exactness at a cost to computation of roughly ten million FLOPS.