Options
2016
Conference Paper
Titel
Systematic evaluation of moving object detection methods for wide area motion imagery
Abstract
Wide area motion imagery (WAMI) facilitates the surveillance of several tens of square kilometers while using only one airborne sensor platform. Typical applications such as automatic behavior recognition, scene understanding, or traffic monitoring depend on precise multiple object tracking. Therefore, moving object detection is generally used as initial step. However, reliable moving object detection for WAMI is challenging as imprecise image alignment, low object resolution and a large number of moving objects lead to split, merged, and missing detections. In the context of this report, a detailed overview of existing methods for moving object detection proposed for WAMI is given. Ten existing methods as well as a novel combination of short-term background subtraction and suppression of image alignment errors by pixel neighborhood consideration are systematically evaluated on the WPAFB 2009 dataset that contains more than 160,000 ground truth detections. Parameters that contribute most to the performance of each method, the influence of related pre-processing steps as well as the impact of varying traffic density and scenery on the performance are discussed.