作者: Florian Pilz , Nicolas Pugeault , Norbert Krüger
DOI: 10.1007/978-3-642-03061-1_14
关键词: Feature (computer vision) 、 Computer science 、 Artificial intelligence 、 Pattern recognition 、 Computer vision 、 Pose 、 Robustness (computer science) 、 Motion estimation 、 Scale-invariant feature transform 、 Rigid body 、 Robotics
摘要: This paper discusses the usage of different image features and their combination in context estimating motion rigid bodies (RBM estimation). From stereo sequences, we extract line at local edges (coded so called multi-modal primitives) as well point (by means SIFT descriptors). All are then matched across time, use these correspondences to estimate RBM by solving 3D-2D pose estimation problem. We test feature sets on various recorded realistic outdoor indoor scenes. evaluate compare results using constraints discuss qualitative advantages disadvantages both types for estimation. also demonstrate an improvement robustness through large data driver assistance robotics domain. In particular, report total failures based only one type relevant sets.