In-Depth Evaluation of Popular Interest Point Detectors on Video Streams

Report ID: 
Steffen Gauglitz and Tobias Höllerer
2009-05-01 05:00:00


We present and in-depth evaluation of popular interest point detectors, which, in contrast to existing evaluations, is targeted towards the application in visual tracking and augmented reality. In particular, candidate algorithms, testbed, and performance criteria are chosen with respect to the application of visual tracking. We evaluate the impact of individual algorithm parameters and present results in terms of repeatability, number of features detected, and computation time. We also describe our method to semi-automatically generate ground truth in detail.


PDF icon 2009-08.pdf