Co-Design of Hardware and Software for Facial Expression Recognition Using Xilinx Zynq SoC

This paper presents the development of an advanced embedded computer vision system tailored for facial expression recognition. The system utilizes the Viola-Jones algorithm for accurate face detection in images, combined with a hybrid approach for feature extraction leveraging Local Binary Pattern (LBP) and Histogram of Oriented Gradient (HOG) features. Aiming at optimizing the performance and reducing the complexity of the system, the Relief algorithm is employed for effective feature selection, focusing on the most significant features to enhance processing time efficiency. A Zynq 7020 SoC platform is elected for system prototyping. This platform effectively integrates an ARM Cortex A9 dual-core communicating with FPGA logic. HW/SW Co-Design is developed starting from the profiling results of the original classification algorithm. This design significantly enhances the system's performance, achieving a real-time classification rate of 27.5 frames per second. This represents an approximately 18 times increase in speed compared to the original program implementation.