Multi-scale car detection and localization using contour fragments

This research features multi-scale car detection and localization. Deformable active contour fragments are represented and learned at multiple scales, which directly represent the shape of cars. Accurate localization is also achieved by learning the variation of the car center position through the deformation of each active contour fragment. Employing Real-Boost for the proposed multi-scale deformable contour fragments provides a very good performance on widely used single scale multi-view and multi-scale multi-view car datasets.