A High-Throughput and Energy-Efficient Hardware Processor for Vision-based Object Detection Systems/Intelligent Automation, Inc.

We propose to design and implement a high-throughput and energy-efficient FPGA-based hardware processor (accelerator) for the deployment of Convolutional Neural Network (CNN) architectures at real-time embedded and resource-bound environments that have low size, weight, power and cost (SWaP-C) requirements. CNNs have been shown tremendous success in vision-based object detection tasks. Two different CNNs including standard CNN and Binarized CNN topologies will be developed for vision-based object detection applications and implemented on FPGA. The proposed solution will be able to process tens of gigapixels per second in real-time SWaP limited environment to support vision-based object detection applications.