Automated photogrammetric method to identify individual painted dogs (Lycaon pictus)

Abstract

The painted dog, Lycaon pictus, has been visually identified by their tricolor patterns in surveys and whilst computerised recognition methods have been used in other species, they have not been used in painted dogs. This study compares results achieved from Hotspotter software against human recognition. Fifteen individual painted dogs in Yokohama Zoo, Japan were photographed using camera-traps and hand-held cameras from October 17–20, 2017. Twenty examinees identified 297 photos visually, and the same images were identified using Hotspotter. In the visual identification, mean accuracy rate was 61.20%, and a mean finish time was 4,840 seconds. At 90.57%, the accuracy rate for Hotspotter was significantly higher, with a mean finish time of 3,168 seconds. This highlights that visual photo-recognition may not be of value for untrained eyes, while software recognition can be useful for this species. For visual identification there was a significant difference in accuracy rates between hand-held cameras and camera-traps whereas for software identification there was no significant difference. This result shows that the accuracy of software identification may be unaffected by the type of photographic device. With software identification there was a significant difference with camera-trap height. This may be because the images of one camera-trap at a lower position became dark due to it being in a shadow.