Vehicle shape approximation from motion for visual traffic surveillance

Abstract
Firstly, the edge segments of the parameterized 3-D polyhedral model were projected from the 3-D scene back into the 2-D image. Then, the matching between 3- D model data and 2-D image data was performed on these 2-D edge segments using the Mahalanobis distance between the attributes of the line segments. Finally, the set with the best correspondence between 3- D model edge segments and 2-D image edge segments was found by using an iterative approach. The matching process of the image data with the vehicle model was carried in the 2-D image environment only, while the intrinsic 3-D information of the vehicle over image sequence was not utilized. This observation motivates us

This publication has 3 references indexed in Scilit: