Use cases

Intelligent intersections

picture1

Use Case Description

For this use case, imec presents EdgeLab, a testbed designed to help researchers and companies evaluate and optimize their machine learning algorithms for real-world edge devices.

What is Edge AI?

Edge AI refers to the execution of artificial intelligence algorithms directly on devices at the “edge” of the network—close to where data is generated—rather than in centralized cloud servers. By processing data locally on cameras, sensors, or embedded devices, Edge AI reduces latency, saves bandwidth, and improves privacy and reliability. This approach is essential for real-time applications such as traffic monitoring, industrial automation, and healthcare, where decisions must be made instantly and securely. EdgeLab provides a unique environment to explore these opportunities by bridging advanced AI models with the constraints and capabilities of real-world edge hardware.

EdgeLab's offering

EdgeLab enables benchmarking across a variety of state-of-the-art edge hardware and AI accelerators, such as the NVIDIA Jetson family or the Axelera Metis. Test the inference of your models remotely for latency, energy efficiency, and accuracy, and receive insights into fine-tuning algorithms for specific devices and data input. Backed by imec’s research expertise in compute and hardware-software co-design, this is your chance to explore what it takes to deploy and run algorithms on edge devices. Whether you're new to AI, a firm believer in the cloud, or an experienced edge practitioner, we have the right formula for you to explore AI technology—and find out if you are ready for the edge.

Key characteristics

  • Traffic data from our sensor clusters
    A diverse array of sensors (e.g., RGB, SWIR, event-based) will be deployed in real-life environments, and this data will be available for inference in EdgeLab. In this way, EdgeLab sets itself apart from traditional benchmarking frameworks, which rely on standardized datasets and workloads and often produce performance indicators that may not accurately reflect your specific use case.

  • State of the art hardware
    EdgeLab provides hands-on access to a broad portfolio of cutting-edge edge devices and AI accelerators. From off-the-shelf platforms like NVIDIA Jetson to new chip designs, participants can experiment with different hardware architectures. This diversity enables developers to assess trade-offs in latency, energy consumption, and accuracy, ensuring that their algorithms are optimized for the right deployment environment.

  • Opportunities for Research and Industry Collaboration
    Researchers and companies are invited to benchmark their algorithms on our edge devices. This collaborative approach fosters mutual learning and provides a platform for participants to refine and showcase their innovations within our testbed.

Results, Highlights & Research Outputs

In 2025 we continued building up the hardware and sensing infrastructure for this use case: four off-the shelf edge devices of the NVIDIA Jetson and Axelera Metis families were deployed, in collaboration with IDLab’s xNative lab. In 2025 the team worked mainly on managing them through Kairos and MaaS, using Kubernetes, which should enable an efficient scale-up to more devices.

On the sensor side, two sensor setups were finalised in 2025: one for data capture on a driving vehicle and one for static data capture on an intersection. In 2025 several data captures were done with the vehicle setup in different urban environments, different weather conditions and different lighting conditions. This data us currently being labelled and available for FAIR researchers. Additionally the first data capture with the static rig happened on an intersection in the city of Mechelen, in collaboration with the citcom.ai TEF. 

Outputs & Resources

Infrastructure / Data Collection

  • EdgeLab testbed.

  • NVIDIA Jetson and Axelera Metis edge devices.

  • Vehicle-based sensor platform for data capture in urban environments.

  • Fixed sensor platform for intelligent-intersection monitoring.

  • Labeled multimodal traffic datasets collected in real-world environments.

Demonstrators / Applications

  • EdgeLab benchmarking environment for AI deployment on edge hardware.

  • Vehicle-based traffic sensing and data collection setup.

  • Intelligent-intersection sensing and monitoring setup.

Collaborative Projects

  • Citcom.ai.

Team & Contact:

Contact

UC Lead: Laure De Cock (imec)

Use Case Lead(s)

Laure De Cock, imec

Research Groups

ISSA - imec; MICAS - KU Leuven; IDLab - UGent; IDLab - UAntwerpen