System of Feature Extraction for Video Pattern Recognition on FPGA

Abstract

A system is proposed for image recognition in a video stream, such as inscriptions, road signs. It is able to recognize the patterns in the complex lighting conditions due to processing the high dynamic range (HDR) signals. The image feature extraction is based on the method, which originated from the Retinex method but is adapted to the HDR images. Due to the new method, the bilateral filter is exchanged to the 2D edge-preserving adaptive filter. The filter output gives information for the feature extraction detectors. The experimental HDR video camera with the feature extraction is built around the Lattice HDR-60 board.