Vortrag vom 16.04.2014

Vortragender: Nicolai Wojke

Titel: Tracking Arbitrary Shaped Objects using Implicit Surface Models

Abstract:

For autonomous robots, awareness of moving objects is a necessity for safe navigation in dynamic environments. This task is generally divided into two stages: (1) object detection using trained classifiers for predetermined object classes such as vehicles and pedestrians and (2) estimation of object dynamics for predicting their future position to avoid collision. In this conventional structure appearance information are solely used in object detection, often based on classifiers that remain unchanged during system runtime.

 

In contrast, the diversity in object appearances as well as the large number of different object classes emerging in real world scenarios advocate motion-based object detection and class-free appearance models that can be learned online. Such models reduce the modelling effort necessary to develop robust object detectors for specific object classes and may guide object reidentification in data association.