BayCEER Workshop 2022

Human-wildlife interactions: The potential of AI and camera trapping

O 3.3 in Afternoon Session

13.10.2022, 14:30-14:45, H 36

Assessing interactions of species in complex natural ecosystems is challenging, but combining camera traps and automated image recognition has rapidly improved data acquisition on wildlife activity patterns lately. However, although most natural systems are highly influenced by human activities such as recreational activities, the potential of using camera traps and automated object detection to monitor human and wildlife activities simultaneously is yet rarely used. We tested an algorithm for automated object detection on camera trap data of three study areas in Bavaria (Bavarian Forest, Fichtelgebirge Mountains, Veldenstein Forest). We evaluated the performance of Microsoft MegaDetector, a deep-learning algorithm for automated object detection, on detecting persons, animals and vehicles on 365163 images from 261 camera trap stations. First results show that manually detected objects strongly correlate with AI-detections (0.62 – 0.92 average correlation estimate) and that detection performance varies between camera trap locations and object classes. Although AI-based automated object classification slightly under- or overestimates visitor and wildlife frequencies, patterns of their activities are well detected. Overall, MegaDetector serves as a reliable as well as time and cost efficient tool for handling big data from camera trap images of human and wildlife activities in natural systems. The resulting data on spatiotemporal behaviour of humans and wildlife allows assessments of their complex interactions.

Automated classification of Eurasien Lynx (Lynx lynx) on a camera trap image.

Keywords: machine learning, camera traps, recreation ecology

Export as iCal: