作者: Pascal Steingrube , Stefan K. Gehrig , Uwe Franke
DOI: 10.1007/978-3-642-04667-4_29
关键词:
摘要: The accuracy of stereo algorithms is commonly assessed by comparing the results against Middlebury database. However, no equivalent data for automotive or robotics applications exist and these are difficult to obtain. We introduce a performance evaluation scheme metrics at three different levels. This can be reproduced with comparably low effort has very few prerequisites. First, disparity images evaluated on pixel level. second level evaluates roughly column column, third performs an object compare real-time capable methods show that global method, semi-global matching, yields best using our incorporate both robustness.