2.1 Image datasets
To examine whether the performance of feature-based image matching
software packages varies for different populations of African wild dogs,
we considered image catalogues from two wild dog populations, one that
spans Laikipia, Samburu, and Isiolo Counties in Kenya (37°2′E, 0°6′N)
and another from the Savé Valley Conservancy in Zimbabwe (32°4’E,
20°3’S). The Kenyan dataset contained images taken between 2004 and 2017
(n = 9139), and the Zimbabwean dataset contained images taken between
2010 and 2013 (n = 2066). In Kenya, these images were taken with ten
different cameras (Olympus© C765UZ, Canon© PowerShot A720IS, EOS Digital
Rebel XT, 10D, 40D, and 60D, Fujifilm© FinePix S5500, Kodak© Easyshare
Z1015IS, Nikon© D70s, and Nikon© Coolpix90). In Zimbabwe, they were
taken with five different cameras (Canon© EOS 450D, 20D, and Digital
Rebel XT, and Panasonic© DMC-FZ20, and Zoran© Coach). Both datasets were
collected as part of long-term monitoring programmes, and contained
images of both single wild dogs and groups of wild dogs, ranging in
their posture from lying down to walking.