A High Throughput Fully Parallel-Pipelined FPGA Accelerator for Dense Cloud Motion Analysis

Tracking of cloud characteristics helps in detecting cyclones which accounts for disastrous winds and heavy landfall on low lying coastal areas. Dense Motion Estimation is one of the most computationally intensive part of any tracking system. In this work we propose an efficient and high throughput hardware implementation of the Dense Horn-Schunck(DHSOF) algorithm to accelerate the cloud motion analysis software running on host PC. A Xilinx Virtex-7 FPGA interfaced to host PC via Peripheral Component Interconnect Express(PCIe) acts as hardware accelerator. The DHSOF architecture can process 3750 x 3750 resolution images in real-time (30fps), which is highest among the state of art methods in the literature. An analysis from the subsystem level to the whole design is performed to find optimal bit width for each stage. Results are verified using standard Middlebury dataset and validated using real world GOES-14 satellite images.