An FPGA-Optimized Architecture of Variational Optical Flow

Abstract

The variational methods for dense optical flow estimation are widely known and adopted techniques for motion detection, object tracking, 3D reconstruction and autonomous navigation in image processing applications. A non-linear model of variational optical flow estimation is the most accurate but also most complicated and computationally intensive method and its implementation in an FPGA is compromise from both a design complexity and a performance. The article is devoted to the specific FPGA-based solution which has been implemented using Verilog hardware description language. The suggested solution is able to process the non-linear optical flow in real time and might be applied as FPGA -accelerator for optical flow processing.