Application of YOLOv4 for Detection and Motion Monitoring of Red Foxes.

Abstract

Simple SummaryThe use of surveillance videos of animals is an important method for monitoring them, as animals often behave differently in the presence of humans. Moreover, the presence of humans can be a source of stress for the animals and can lead to changes in behavior. Extensive video material of red foxes has been recorded as part of a vaccine study. Since manual analysis of videos is both time-consuming and costly, we performed an analysis using a computer vision application in the present study. This made it possible to automatically analyze the videos and monitor animal activity and residency patterns without human interference. In this study, we used the computer vision architecture ‘you only look once’ version 4 (YOLOv4) to detect foxes and monitor their movement and, thus, their activity. Computer vision thereby outperforms manual and sensor-based exhaustive monitoring of the animals.Animal activity is an indicator for its welfare and manual observation is time and cost intensive. To this end, automatic detection and monitoring of live captive animals is of major importance for assessing animal activity, and, thereby, allowing for early recognition of changes indicative for diseases and animal welfare issues. We demonstrate that machine learning methods can provide a gap-less monitoring of red foxes in an experimental lab-setting, including a classification into activity patterns. Therefore, bounding boxes are used to measure fox movements, and, thus, the activity level of the animals. We use computer vision, being a non-invasive method for the automatic monitoring of foxes. More specifically, we train the existing algorithm ‘you only look once’ version 4 (YOLOv4) to detect foxes, and the trained classifier is applied to video data of an experiment involving foxes. As we show, computer evaluation outperforms other evaluation methods. Application of automatic detection of foxes can be used for detecting different movement patterns. These, in turn, can be used for animal behavioral analysis and, thus, animal welfare monitoring. Once established for a specific animal species, such systems could be used for animal monitoring in real-time under experimental conditions, or other areas of animal husbandry.