Vote3Deep
Transferring tech from lab to vehicle
for high performance, real time object detection
Project Outline & Objectives
The project intends to assess the feasibility of transferring a deep-learning research solution,
Vote3Deep, from server grade to portable systems ready for integration in connected and autonomous vehicles.
This is required since to enable autonomy. We need fast and accurate detection of objects.
Objectives:
To achieve commercially viable solution, software must be able to run in real-time with extremely high accuracy, on low cost hardware.
The translation of this research solution into a commercially viable product will be enabled through a detailed test programme at the UKAEA owned Culham Science Centre, where it will be trialled extensively both statically and in-car in a secure environment.
Challenges:
Down-costing hardware without a loss in performance.
GPU optimisation.
Integration into a vehicle, with the software running in real-time.
Successes:
Object detection software package transferred from high-cost server to low-cost portable hardware.
No loss in performance or accuracy.
Integration into a vehicle, with the software running in real-time.
Milestones:
Vote3Deep concept proven on portable system
Successfully integrated with tracking system
Vote3deep prototype proven on portable system
Meets performance targets in validation scenario
Development pathway defined.