Drinea (DEVEX)

The CIDN–DRINEA platform as a whole brings together 10 project fact sheets with the following objectives:

• Structuring a Human–Machine Collaboration ecosystem in Normandy
• Developing a single interface capable of simultaneously coordinating drones and rovers to assist operators in complex contexts

Target Sectors:

• Defense
• Civil security
• Environmental protection

CESI is responsible for one project entitled DEVEX – “Vehicle / Human Detection in Outdoor Environments”, whose objective is to challenge object detection strategies developed in a context different from that of an industrial manufacturing workshop, namely complex outdoor environments, for example in military applications. These two types of environments share many similarities, one of the main ones being the lack of annotated data.

Achievements as of January 13, 2026

• Participation in the COHOMA III Challenge to develop the AI building block for target detection using mixed datasets
• Continued scientific work by comparing different dataset creation strategies and evaluating their performance on the YOLOv11 deep learning detection model:
  • Mixed: limited quantities of real data + synthetic data generated from the Digital Twin (DT)
  • Augmented: limited quantities of real data + augmented synthetic images (real images + CAD object models). Use of Glomap – SfM to extract scene planes from point clouds
  • Randomized: limited quantities of real data + augmented synthetic images applying randomization techniques (texture variations, lighting, etc.)
• Currently, the focus is on industrial objects, leveraging those available in the flexible production workshop

Perspectives

• Reproduce the same scientific approach for the Defense context: analyze whether the trends observed in industrial object recognition are also valid in military outdoor environments, for other object categories and for people
• Evaluate additional detection models: RF-DETR (Roboflow), Fast R-CNN, lightweight models, etc.