Fast Learning and Detection of Edge Shapes.

Walterio Mayol-Cuevas (University of Bristol, UK)

Abstract:

Shape representation and detection is a crucial problem in Computer Vision. In general, finding a configuration of edges on another image is an un-tractable problem that underpins methods that attempt to work with those pervasive texture-less objects around us. Together with the importance of fast shape-based object detection, is the importance to be able to quickly capture and learn those objects.

In this talk, we will present a method developed to address these two issues. It uses a very compact and fast to compute representation of edge shapes and is amenable to work in low-end mobile hardware for dozens and or hundreds of multiple-object views. We will also discuss the application of this method within our ongoing work in egocentric perception.