Drone.AI project is integrating a set of technical and operational assets to deliver a quick, automated and field deployed solution.
DroneAI solution, as an end-to-end solution, will provide two major capacities:
_ a local data processing chain that will apply the AI models on the captured images following the requests of the field personnel,
_ a central solution allowing data-scientists or GIS specialist to create and distribute AI models.
The solution will be integrated into the emergency.lu mobile, satellite-based, telecommunication platform.
A full stack solution
DroneAI solution relies on the Max-ICS platform developed and operated by EarthLab Luxembourg. Drone.AI is using the Deep Learning factory (C) concept allowing to easily create, train, test and deploy machine and deep learning models. Max-ICS also provides a modern Platform as a Service solution covering all the requirements in term of application hosting, deployment, security and scaling.
AI on the edge
Thanks to the Max-ICS platform, DroneAI offers a complete coverage from the training to the inference of AI models.
Artificial Intelligence inference operations are very demanding in term of processing ressources and these inference operations are here performed on the field of operation, close to the capture of the observation information. To maintain a good level of performance, a dedicated edge component, relying on the nVIDIA AGX Xavier hardware, is integrated into the NoSaCo devices built by HITEC Luxembourg and used as terminals within emergency.lu platform. Max-ICS platform is managing this device performing automatically the maintenance operations, the configuration of the device avoiding complex IT operations from the field.
Large variety of AI operations
Max-ICS platform offers a large variety of computer vision tasks and trained models. In the context of the Drone-AI project, different models are already accessible to demonstrate the large capabilities of the solution. For example, models to detect the roof damages, to look for road destruction or building vulnerability detections, are directly accessible.
It is also possible to build new models or extend the existing ones using the model training features accessible from the Max-ICS platform.