Fast Keypoint Recognition and Recent Work at CVLab

Опубликовано 6 сентября 2016, 17:59

I will first present our approach to real-timekeypoint recognition based on a Naïve Bayesian classification framework, and its applications to robust 3D object detection and tracking. I will also discuss a recent extension to the estimation of the keypoint perspective orientations, suitable for detecting low-textured objects. Then, I will quickly present more recent developments at CVLab, in particular deformable surfaces registration in closed-form, dense local descriptors for wide baseline reconstruction, and accurate monocular 3D tracking using specular reflections.