| Popis: |
This paper describes a method for omni-directional 3D reconstruction with a hyperbolic mirror and a regular camera. Omni-directional images are becoming increasingly famous in computer vision and other fields. Despite there some related research in omni-directional vision system applications during last two decades, however, omni-directional perception system have not ever been applied in crime scene reconstruction. The main advantages of the catadioptric vision system are robustness because of no needing mechanical moving units and single effective viewpoint constraint which allow the next feature detection. Firstly, we provide the transformation mapping method from the omni-directional images into the cylindrical panoramic images. Secondly, we use the SIFT Method to yield the scenario features. Subsequently, we tackle this matching and disparity estimation problem by proposing Markov random field (MRF) model. Finally with some experiment and we show the 3D clouds actually results. |
