Memuat
Memuat
Software

KarhutlaVision monitors watchtowers, concession perimeters, peatland areas, patrol routes, and fire brigade posts/command posts in real time using AI computer vision (YOLO) to detect smoke, hotspots, and high-risk activity -- triggering alerts to field teams within seconds instead of waiting hours for manual reports.
Unlike conventional CCTV systems, KarhutlaVision runs portably on site (edge deployment), so it keeps working even with limited internet connectivity -- ideal for watchtowers in remote locations that may not have stable bandwidth.
Normal Conditions, Light Smoke, Thick Smoke, Fire Spot/Open Flames, Embers/Hotspots, Spreading Smoke Haze, Land Burning Activity, Unauthorized Vehicles/Personnel in the Concession Area, Personnel Without PPE in Hazard Zones, and Fire Brigade Units in Operation -- each mapped to a severity level (high/medium/normal) and an operational category (Fire, Early Warning, Air Quality, Compliance, Security, Safety, Operations).
1–3 of 17
best.pt as the active modelThe YOLO model is trained specifically for watchtower and forest/land concession scenarios (including peatland) -- not a generic model -- with 10 detection classes mapped directly to real operational and regulatory compliance risks. National forest and land fire penalties imposed on dozens of companies can reach hundreds of billions of rupiah per company, making early detection not just an operational tool but a financial and legal risk mitigation measure. Available as an Outright Purchase, Rental/Subscription, or Hybrid scheme to suit your company's budget.
Ideal for industrial plantation forest (HTI) and plantation concession (HGU) holders, peatland concessions, and forestry/plantation groups with more than one watchtower that want to move away from manual supervision and paper-based reporting.
Built with Python (Flask + Flask-SocketIO) and a YOLO object detection model (Ultralytics/PyTorch), running portably on-premise/at the edge at watchtowers or command posts -- with no dependence on third-party cloud infrastructure, making it ideal for locations with limited internet bandwidth. Automatic installation via a single launcher file that detects available Python and NVIDIA GPU (CUDA). Report exports support PDF, Excel, and Word formats.