This conference poster presents WatchEDGE smart cameras for real-time wildlife detection at the far edge. Animals can severely affect crops, making continuous monitoring important for both farm management and wildlife protection. The work investigates deep-learning-based animal detection on resource-constrained far-edge devices, including GPU-equipped trap-camera systems, under limited compute, power, and rural network connectivity.The poster introduces additional on-the-fly multi–color-space transformations into the YOLO Ultralytics training pipeline, complementing existing augmentations such as mosaic. These chroma/luma parameterizations are designed to improve robustness to domain shifts, camera sensor effects, and evening-time/adaptive imaging conditions while supporting continued inference on far-edge devices.The deployment architecture uses a custom battery-powered camera system at the far-edge site. The camera connects to local customer-premises equipment over WLAN through a Wi-Fi extender, while the far-edge site communicates with the edge over a 4G cellular network. Instead of transmitting full video sections, semantic detection information is sent to the dashboard to reduce bandwidth usage.The poster reports edge-oriented validation using an RTX 3050 laptop GPU and a Jetson Orin Nano, comparing baseline YOLO models with the augmented YOLO-AUG pipeline for wildlife detection.Presented as a conference poster at the 22nd Italian Networking Workshop (INW), Bardonecchia, Torino, Italy, 12–14 January 2026.
Jean Pierre Asdikian, Sebastian Troìa, Mengyao Li et al.· Figshare· 0 citations
This conference poster presents WatchEDGE smart cameras for real-time wildlife detection at the far edge. Animals can severely affect crops, making continuous monitoring important for both farm management and wildlife protection. The work investigates deep-learning-based animal detection on resource-constrained far-edge devices, including GPU-equipped trap-camera systems, under limited compute, power, and rural network connectivity.The poster introduces additional on-the-fly multi–color-space transformations into the YOLO Ultralytics training pipeline, complementing existing augmentations such as mosaic. These chroma/luma parameterizations are designed to improve robustness to domain shifts, camera sensor effects, and evening-time/adaptive imaging conditions while supporting continued inference on far-edge devices.The deployment architecture uses a custom battery-powered camera system at the far-edge site. The camera connects to local customer-premises equipment over WLAN through a Wi-Fi extender, while the far-edge site communicates with the edge over a 4G cellular network. Instead of transmitting full video sections, semantic detection information is sent to the dashboard to reduce bandwidth usage.The poster reports edge-oriented validation using an RTX 3050 laptop GPU and a Jetson Orin Nano, comparing baseline YOLO models with the augmented YOLO-AUG pipeline for wildlife detection.Presented as a conference poster at the 22nd Italian Networking Workshop (INW), Bardonecchia, Torino, Italy, 12–14 January 2026.
Jean Pierre Asdikian, Sebastian Troìa, Mengyao Li et al.· Figshare· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
