Application of YOLO Object Detector for Horse Stress Monitoring During Transportation

The emerging field of ubiquitous computing leads to new application and processing speed requirements in various every day aspects. When it comes to image processing tasks, which require many computations, the limited resources of embedded systems make those tasks rather challenging. This paper explores an implementation of an object detection task deployed on an embedded device NVIDIA Jetson Nano GPGPU: YOLO object detector detects the nostrils of horses using an infrared camera during transportation. Results of the proposed algorithm can directly be used for calculating the breathing rate and to find out if the horse is currently under stress. We compared YOLOv5 and YOLOv8 as object detection models. In the proposed implementation on the NVIDIA Jetson Nano GPGPU the inference time is around 40ms, which fulfills realtime processing requirements for the given environment. Both YOLO versions tested show high accuracy, while the additional robustness tests indicate a slight advantage of YOLOv5 comparing to YOLOv8.