ICARUS UAS Detection

Model Overview

Detect, localize, and counter drones in diverse environments with the ICARUS UAS Detection model, a deep learning-based counter-drone detection model performing from a ground-to-aerial perspective. Trained using a wide collection of photorealistic 3D models and environments, ICARUS is able to detect a variety of drone types including quadcopters, octocopters, hexacopters, and fixed-wing drones.

ICARUS was trained in various environmental conditions including fog, rain, snow, sun glare, and low lighting, as well as challenge scenarios including distance or small objects down to 93 POT, multiple targets, and sensor degradation. This algorithm was developed to perform for both cloud-based inference and drone-mounted sensors, and is effective on existing camera systems including low cost devices.



Optimized for edge performance with a low SWaP


Effective for a variety of UAS', including fixed wing drones


Ground-to-aerial perspective


Trained to perform from dawn to dusk


Effective for small and distant objects down to 93 POT


CVEDIA is a pioneer in the use of synthetic data for machine perception. We use deep learning to create synthetic-based detection and classification algorithms. CVEDIA models actually outperform traditional models because synthetic data allows for feature-based design. That means our models are safer and quicker to market. Today we work with over 30 of the world’s largest companies on their most complex deep learning projects.

Every model is backed by an ongoing maintenance service, as well as the ability to customize and add to a model through our in-house development team. We’ll even help you deploy our models in your software stack. Contact us for more information.

Operating Specifications




Visible light


Backbone: Resnet-101, Detector: RefineDet



CVEDIA Services


Each CVEDIA model comes backed with an ongoing maintenance agreement. Our team is available for continuous improvements and to ensure upkeep. Reach out to us for more information.

Fine Grain Classification

Speak to our team to discuss model classification as a service add on or as a separate model. CVEDIA’s in-house team will work with you to define your requirements, and build a synthetic-based model incorporating your drone classification specifications.

Modelling Another Use Case or Perspective

Looking for a UAS Detection algorithm from another perspective, or for a different sensor type? Contact us to discuss a custom solution.

Use Cases


Use the ICARUS UAS Detection algorithm as a part of a larger security framework. Enhance your existing security grid and ensure your system is ready for anything with a model trained using a wide variety of drone makes and models.


Maintain an airport landing space free of foreign objects with the ICARUS UAS Detector, trained extensively with drone model variants including quadcopters, octocopters, and hexacopters.

Perimeter Security

Add to existing perimeter security measures with a model built to detect hundreds of drone model variations, including package delivery. ICARUS has been tailored to reduce false positive in typical air conditions through CVEDIA’s pioneering use of synthetic data.

Public Events

Ensure peace of mind in public spaces and events with less resources using the ICARUS UAS Detector. CVEDIA models help you surpass the data collection and training phases of your project, so that you can get to market quicker.

Read the Performance Report

CVEDIA provides benchmarked performance reports for every model – they’re the first of their kind. Designed by data scientists, the ICARUS UAS Detector report includes thorough statistics on performance in various conditions and environments, as well as information on uses and limitations. Click below to receive it.