FPGA-based Implementation of HOG Algorithm: Techniques and Challenges

Abstract

Histogram of Oriented Gradients (HOG) is a method for extracting features from an image, which has many applications in Computer Vision. Due to the complexity and high amount of computations of this algorithm, software-based implementations of HOG cannot meet the real-time criterion. Therefore, many researchers have implemented HOG algorithm on hardware platforms such as FPGAs. This paper presents an extensive review of FPGA-based implementations of the HOG algorithm, that have been published from 2010 to 2019. Different techniques for hardware implementation of HOG are classified into three groups: methods which improve a certain stage of the algorithm, methods which optimize the whole algorithm, and methods which make minor simplification on the algorithm. In this paper, these three classes of techniques are reviewed. Finally, the speed and resource utilization of the surveyed papers are compared to each other in order to present a comprehensive conclusion on FPGA-based HOG implementation.