FIRA-DAV: Feature-level Image and Radar Fusion for Detect and Avoid

Fri, 01 Sep 2023 00:00:00 +0000

Project Overview

FIRA-DAV (Feature-level Image and Radar Fusion for Improved Detect and Avoid Functionality in Autonomous Aerial Vehicles) is a Proof-of-Concept project funded by the Athena Research Center Technology Transfer Office under the internal PoC programme (ΔΕΜΤ, project No. 80260). The project was competitively selected in June 2023 following evaluation by a three-member expert panel, and ran from September 2023 for 6 months.

As Principal Investigator, I led the design, equipment procurement, and dataset construction for radar–camera sensor fusion targeting UAV Detect-and-Avoid (DAA) — a critical safety requirement for beyond-visual-line-of-sight (BVLOS) operations. The project was carried out in collaboration with Hellenic Drones (industry partner, Dr. Christos Skliros) and ROBOSURVEY LTD (equipment supplier).

Team

Dr. Evangelos Vlachos — PI, Athena Research Center / ISI
Hellenic Drones (Dr. Christos Skliros) — Industry partner

Equipment & Dataset

The project procured a Texas Instruments IWR6843AoP 77 GHz FMCW mmWave radar (Antenna-on-Package variant) and a synchronized CSI camera, integrated on a drone-mounted embedded platform. Measurement campaigns produced a synchronized dataset of radar 3D point clouds and RGB camera frames in real flight conditions, including UAV-to-UAV scenarios for obstacle detection.

Key technical work:

Data collection: Synchronized radar point clouds and RGB frames captured in real UAV flight scenarios
Calibration: Radar-to-camera extrinsic transformation, intrinsic camera calibration
CFAR detection: Configuration of detection and tracking layers for the IWR6843AoP sensor
Applied for Google Cloud Research Credits to support deep neural network training on the collected dataset

Technical Outcomes

Synchronized radar–camera dataset collected under real flight conditions
Derivation of transformation function mapping radar 3D point cloud to 2D image plane
Training pipeline for spatial attention fusion neural network combining radar and vision features
Efficient Python and Julia code for embedded AI deployment
Dataset and equipment directly adopted by EUSOME (Horizon Europe, 2025–2028), which carries no equipment budget

Reviewer Recognition

The proposal was rated positively by all three external reviewers:

“This is a huge market and everyone is working on it. This tech, if unique and solves real issues, has cross applicability.”

“Deep-tech IT solution that relies on state-of-the-art technology and strong know-how that is probably difficult to be copied.”