The Frame Image Offering Tracker Algorithm

14 Copyright © 2012 Open Geospatial Consortium.

6.5 The Frame Image Offering

The Frame Image Offering was implemented as a method for retrieving a raw jpeg image for a frame from the motion imagery using a time instant temporal filter. This offering fulfilled the same functions as the video offering, but for a single frame. This offering was used by client software to retrieve video imagery, but was not used by the tracking algorithm. 7 WPS Tracker implementation

7.1 Tracker Algorithm

In order to turn raw video and VMTI detections into a moving object track, a processor, typically known as a tracker is involved. For OWS-8, this processor was implemented using the OGC WPS 1.0.0 specification Yu Chen, 2011. The algorithm for detection- to-track is based on the Kanade-Lucas-Tomasi feature tracker. All frames, not just the frames with metadata, must be retrieved together in order to better facilitate the assumption of the tracking algorithm. The algorithm assumes that an object moves only slightly across frames. If there is a big jump of tracking object between frames, the algorithm would not work or easily lose the object under tracking. Once the feature is found in the target frame, the closest detection from the metadata is used in the track if there is one such detection point in the metadata that is not farther than 10 pixels in this case. KLT also supports multi-resolution tracking to deal with relatively large displacements of targets between video frames. Affine computation was enabled with KLT to deal with non-consecutive frames. In the implementation, empirical adaptation and adjustment were supported to achieve best results of tracking in a geospatially-aware sense. These include enhancements for : ฀ being adaptive to the size of target under investigation by defining the tracking target window size, ฀ being capable of varied range of displacements by adaptively adjusting the parameters for multi-resolution pyramid, and ฀ being tolerant to occlusion by using a momentum to track the target through occluded area. Copyright © 2012 Open Geospatial Consortium. 15

7.2 Tracker Inputs and Execution Sequence