Wildfires are a recurring threat in Portugal, particularly during the summer months. The speed at which they spread makes early detection crucial, yet traditional monitoring methods often fall short. This challenge led us to start using machine learning models capable of detecting fire and smoke from aerial imagery captured by our UAVs. Initial testing has been promising, demonstrating the potential for AI-driven wildfire detection to enhance response times and minimize damage.
Beyond fire detection, our team is deeply invested in research and development (R&D) across multiple domains. From pioneering new manufacturing techniques to identifying novel applications for UAVs, we are continuously pushing the boundaries of what our aircraft can achieve. A key area of focus is autonomy—reducing the need for human intervention and enabling our UAVs to operate seamlessly in complex environments.
Overcoming the Challenge of Latency
From the outset, we chose Ardupilot as our autopilot system due to its powerful feature set, open-source nature and extensive customizability. Our UAVs rely heavily on Pixhawk flight controllers, which have consistently delivered outstanding performance. Redundancy is a critical factor in our design philosophy, ensuring system reliability, while the ability to use standard connectors simplifies integration and maintenance.
Historically, one of the biggest challenges in UAV-based wildfire detection has been processing aerial footage in real time.
The Solution: Onboard AI with the Pixhawk-Jetson Baseboard
The key to overcoming these challenges lies in onboard processing. Running AI models directly on the UAV eliminates the significant delay stemming from the need to transmit video over long distances, significantly reducing latency and improving responsiveness. However, most embedded computers are either too weak to handle real-time inference or too heavy, impacting flight efficiency.
This is where the Holybro Pixhawk-Jetson baseboard comes into play. By integrating a Pixhawk flight controller with an NVIDIA Jetson Orin Nano, it combines robust flight control with powerful AI capabilities. This allows us to process video onboard, detect fires in real time, and make intelligent flight decisions autonomously—all without compromising performance.
Image processing comparison: Ground-based (left) with up to 5s latency vs. onboard (right) with just 100ms, enabling faster decision-making.
Looking Ahead
With these advancements, we are making significant strides towards smarter, more autonomous UAVs for wildfire detection and beyond. The possibilities extend far beyond emergency response—agriculture, environmental monitoring, and infrastructure inspection could all benefit from similar onboard AI systems.
We’d love to hear your thoughts: What other applications do you see for onboard AI in UAVs? If you’re working on similar challenges, let’s connect and share insights!
More updates coming soon—stay tuned!
https://aerotec.pt/atlas
Related
Discover more from sUAS News
Subscribe to get the latest posts sent to your email.