CORDIC-based Parameters-fusion HOG IP For Extracting Feature

抄録

Histograms of Oriented Gradients (HOG) descriptor significantly outperforms other features for object detection, especially for pedestrian detection. Exploitation of high performance and efficient HOG IP has been a research hotspot in Automotive Electronic. In this letter, a CORDIC-based Parameters-fusion LUT HOG algorithm is proposed and made as the hardware IP (Intelligent Property). To decrease the computational effort and improve the accuracy, pipeline Coordinate Rotation Digital Computer (CORDIC) is applied to generate the magnitude and orientation of gradient. Parameters- fusion Look-Up-Table (LUT) based regional division can operate the tri-linear interpolation in HOG with high speed. Circuit design and chip fabrication were performed using 0.18 μm CMOS technology. Measurement result shows that this CORDIC-based Parameters-fusion LUT HOG IP reduces the hardware overloads and can be used as the general IP for extracting features.