作者: Tobi Vaudrey , Clemens Rabe , Reinhard Klette , James Milburn
DOI: 10.1109/IVCNZ.2008.4762133
关键词:
摘要: Performance evaluation of stereo or motion analysis techniques is commonly done either on synthetic data where the ground truth can be calculated from ray-tracing principals, engineered easy to estimate. Furthermore, these scenes are usually only shown in a very short sequence images. This paper shows why may not testing criteria by giving evidence conflicting results disparity and optical flow estimation for real-world testing. The dealt with this images taken moving vehicle. Each contains 250 image pairs more. Synthetic driver assistance (with truth) 100 more pairs. Particular emphasis paid scene sequences. All used made publicly available at http: //www.mi.auckland.ac.nz/EISATS.