cREAtIve targets the development of novel highly-adaptable embedded deep learning solutions for automotive and traffic monitoring applications, including position sensor processing, scene interpretation based on LIDAR, and object detection and classification in thermal images for traffic camera systems. These applications share the need for deep learning solutions tailored for deployment on embedded devices with limited resources and featuring high adaptability and robustness to changing environmental conditions. cREAtIve will develop knowledge, tools and methods that enable hardware-efficient, adaptable and robust deep learning.
Deep learning applications without the cloud
Deep learning solutions for object classification rely mostly on cloud-based data processing which has sufficient power, bandwidth and storage to operate.
Camera-based systems can also use deep learning algorithms to detect, recognize and track objects, but these systems need zero latency and are impacted by privacy concerns. Since they cannot rely on the processing power of the cloud, these systems must be equipped with embedded hardware that is able to perform efficient and robust deep learning algorithms in real-time.
Measurements provided by position sensors are corrupted by non-stationary noise caused by varying environmental conditions. Novel lightweight noise canceling solutions capable to cope with the non-stationary nature of the perturbations and operating directly on the sensing devices are needed in order to increase the accuracy and robustness of the sensor measurements.
LIDAR (laser-based detection) systems are essential in advanced driver-assistance systems and autonomous driving. Innovative embedded deep learning solutions are required in order to improve object detection, tracking and classification accuracy for such devices.
In addition to coping with power, bandwidth and processing constraints, a camera-based deep learning solution for traffic management would also need to adapt to different environments and weather conditions. The cREAtIve consortium will develop an original deep learning solution with the required adaptability, as well as the underlying hardware needed to optimally support it.
Embedded devices become intelligent
cREAtIve consortium partners are experts in thermal cameras and traffic monitoring solutions, automotive sensors, LIDAR systems, deep learning and adaptable hardware solutions. Together, they will address the following innovation goals:
- Develop AI sensors that filter out noise caused by temperature changes and stray magnetic fields;
- Improve the processing of LIDAR (laser-based detection) measurements to boost 3D object detection and tracking performance;
- Design a new deep-learning paradigm that enhances the accuracy of object detection and classification in thermal images;
- Investigate hardware trade-offs for deep networks in embedded and edge systems where resources and power consumption are constrained;
- Empower the embedded deep network to adjust itself to different environments.
Improved traffic safety and more
The outcomes of the cREAtIve project will be key advancements in the development of competitive, adaptive AI solutions in resource-limited embedded devices. In practice, the results of the project could be used to provide stable measurements in position sensing devices in automotive industry, better vehicle detection and tracking in autonomous driving, and improved traffic management and pedestrian counting.
“The cREAtIve project will develop novel, adaptive and robust deep learning solutions for resource-limited embedded devices.”
Reconfigurable Embedded Artificial Intelligence
cREAtIve is an imec.icon research project funded by imec and Agentschap Innoveren & Ondernemen.
It started on 01.04.2019 and is set to run until 31.03.2021.Download as pdf
- Flir Systems Trading Belgium
- imec - IDLab Data Science Lab - UGent
- UGent - Hardware and Embedded Systems (HES)
- VUB - Vakgroep Elektronica en Informatica (ETRO)
- Project lead: Wouter Favoreel
- Research lead: Joni Dambre
- Proposal Manager: Poona Bahrebar
- Innovation manager: Eric Moons