Scene recognition based on gradient feature for autonomous mobile robot and its FPGA implementation

Abstract

This paper introduces the image processing system of scene recognition using gradient feature and its FPGA implementation for mobile robots usage. We propose a hierarchical gradient feature descriptor, which could be easily implemented in a compact size of logic circuit on the FPGA. The gradient feature includes the function of corner detection by using the dispersion of directional gradient. The proposed hierarchical gradient feature analyzes the magnitude and direction in 17 regional blocks, where the input image is smoothed by the 7 line buffers of Gaussian filter with 8 parallel circuits as preprocessing.