| Submitted to VS-PETS, October 2005 (2005) | |||||||||||||
Abstract | |||||||||||||
| This article compares the performance of target detectors based on adaptive background differencing algorithms on public benchmark data. Several state of the art methods are described together with their parameterization. The performance is evaluated using computation time, recall and precision with respect to annotated ground truth. A surprising result is that the method with the most complex background model is outperformed by an efficient alternative implementation and a very simple background model combined with a Kalman filter. | |||||||||||||
Publication details | |||||||||||||
| |||||||||||||