Embedded sensor fusion network

Highly accurate 3D object detectors require significant computational resources, and reducing computation and memory load while maintaining the same level of performance is a critical task for any safe and reliable autonomous vehicle. This research project investigates the deployment of an accurate 3D object detection model to a resource constrained architecture by changing the model structure, its parameters as well as its activity during operation. Through a multi-level optimization, both the amount of computation as well as the memory load will be reduced while maintaining 3D object detection performance. The knowledge gained from this experiment will help the industry partner to develop new architectures for ever more complicated computational tasks.

Intern: 
Rytis Verbickas;Yahya Massoud
Faculty Supervisor: 
Robert Laganiere
Province: 
Ontario
Partner: 
Partner University: 
Program: