(1) DNN training, optimization and visual monitoring algorithms for visual intelligence applications (2) DNN simulator and development tool software (3) Functional and architecture specifications of DNN vision accelerator (4) FPGA implementation of DNN accelerator (5) System specification of FPGA SoC (6) Firmware for semantic information processing and application on FPGA SoC (7) FPGA SoC evaluation board and system (8) Reference design of real-time monitoring camera (CS deliverable for Customer #1) (9) Customized System on Chip (“SoC”) Field-Programmable Gate Array (“FPGA”) for video analysis (CS deliverable for Customer #1) (10) Software for Infant visual monitoring system (in binary format) (CS deliverable for Customer #2)
Manford Development Limited
Zhuhai Huichen Integrated Circuit Design Co., Ltd.
Objective: Deep learning technology ignites a new wave of visual intelligence applications in security and safety, automation, robotics, ADAS (advanced driver-assistance system) and smart retail. This project aims to develop a dedicated SoC platform including hardware accelerator, firmware, and software tools to realize DNN (deep learning neural network) based visual intelligence devices for intelligent video surveillance (IVS) and relevant applications. Another objective is to engage target customers for mass-adoption in the next step. R&D Methodology: In this project, a many-core data flow processing architecture is proposed to efficiently compute DNNs of varied structures and complexities. A dedicated hardware pre-processing module is developed to accelerate the multi-object recognition in high-resolution video input. The firmware realizes semantic information processing and reference applications. In addition, the offline training and optimization tools are developed to train DNN models for various applications. All these technologies will be realized and verified on a SoC FPGA system. Impact and Benefits: The deliverables will enable deep learning based visual intelligence devices for various applications. The mass adoption of these devices will enhance the competitiveness of video surveillance, smart city and smart retail industries.