DECEMBER 15, 1998, 1:00
Vasanth Philomin

CFAR, University of Maryland
Real-time Generic object detection with applications in intelligent vision based navigation

ABSTRACT

In this talk, I will present an efficient shape-based object detection method using Distance Transforms (DTs). The proposed method extends previous DT-based matching techniques by using multiple features and a template hierarchy combined with a coarse-to-fine search over the template transformation parameters. This results in various speed-up factors, which brings a number of matching applications that previously required special-purpose correlation hardware onto the realm of the PC. I will also present results on real-time traffic sign detection to illustrate this approach and show how to extend the method to detect highly non-rigid objects in complex scenes such as pedestrians in an urban scenario. Finally, I will describe an example-based RBF framework that uses intensity cues for the classification of traffic signs and show a video of a prototype "Traffic-sign Assistant" on-board an E-class Mercedes Benz. This talk is a rehearsal for my preliminary exam and so I would appreiciate some feedback from people who attend this talk.