Dynamic management of a partial reconfigurable hardware architecture for pedestrian detection in regions of interest

In this paper, we propose a video streaming architecture implementing an IP core design based on Histograms of Oriented Gradients (HOG) algorithm, by using a high-level synthesis workflow. Also, we present a decision algorithm for a dynamically reconfigurable architecture, applied to people detection in video over a Region Of Interest (ROI). Our method dynamically manages the resource consumption in our hardware accelerator. In the test performed with only one reconfigurable module, our method achieves a power reduction up to 7% and a speed of 51 fps, but these results could be improved according to the number of reconfigurable modules. For implementation, we used a Xilinx Zynq 7000 series platform and our experimental results are compared with other software and hardware implementations. Our hardware architecture can be used in a vast number of vision systems, including advanced driver assistance systems (ADAS).