Simplified histograms of oriented gradient features extraction algorithm for the hardware implementation

Abstract

The Histograms of Oriented Gradient features are common object detection feature descriptors in computer vision and image processing. The features are widely used in the field of pedestrian detection, but the calculation speed is slow, and it is not fit for the situation of high real-time demand. This paper introduces the original HOG feature extraction algorithm, and simplifies some time consuming steps for the hardware implementation, namely gradient calculation in the HOG feature extraction and histogram calculation to meet the requirements for hardware implementation. This paper also particularly analyzed the size of gradient used in different simplified methods, shows the simulation results in Matlab2014a, and compared the calculation results and computing time of different methods, proving the correctness and efficiency of the simplified method, which lays a theoretical basis for the follow-up of hardware implementation.