Abstract:
Hardware based human detection plays a key role in several applications such as smart vehicles, robotics and surveillance. In this paper, an efficient scheme for combined histogram of oriented gradients (HOG) and local binary pattern (LBP) feature extraction on field programmable gate array (FPGA) is presented. HOG and LBP computations for each pixel value of 3 × 3 pixels, 8 × 8 pixels and 16 × 16 pixels regions were performed for the input images. HOG-LBP feature extraction was implemented using both the shape and texture information for more robust human detection in the presence of partial occlusion and background clutter. The proposed scheme uses 793 lookup tables (LUTs) on Virtex-5 FPGA to achieve 82.118 MHz clock frequency. The scheme can be employed for real-time human detection with low power consumption, low cost and real-time performance.