|Project Title:||FPGA based Edge Computing AI System for Video Analytic in Transportation Applications|
|Hosting Institution:||LSCM R&D Centre (LSCM)|
|Abstract:||Intelligent transportation system (ITS) using AI video analytics plays an increasingly
important role in public transport management.
Based on the real time acquisition of urban road traffic flow information and event
recognition, ITS provides intelligent guidance for relieving traffic jams, emergency
handling and traffic control. CNN (convolution neural network) based models, such as
YOLO, are usually used to build ITS detectors.
In this project, LSCM proposes to implement hardware accelerator for convolutional
neural network (CNN) used in ITS using FPGA based technology. We will develop the
solution to take video input and output object or event detection results in a data pipe,
with ITS model implemented in FPGA configurations. The convolutional layers are fully
pipelined for hardware acceleration. Furthermore, the hardware configuration will be fully
parameterized, so that different numerical calculations are mapped to a dedicated
hardware blocks. Process element (PE) will be optimized of floating-point and fixed-point
calculations running in parallel, to achieve higher overall speed. We target to achieve
better speed and efficiency than current deployed hardware platform based on GPU.
|Project Coordinator:||Mr Martin Chun-Wai LAI|
|Approved Funding Amount:||HK$ 2.79M|
|Project Period:||29 June 2021 - 28 Oct 2022|