Design-and-Implementation-of-Face-Recognition-based-on-PYNQ

                                     Design and Implementation of Face Recognition based on PYNQ 
                                                             
                                                             Abstract 
In recent years, face recognition has been widely used in areas such as payment, security, and robotics, and has become a research hotspot in the field of computer vision. Face recognition requires steps such as detection, alignment, and recognition, but traditional methods such as Haar cascade classifiers, PCA do not perform well in detection accuracy and robustness. With the development of deep learning technology, deep learning algorithms are applied to face recognition, which effectively improves the accuracy and robustness of face detection. Traditional algorithms combined with deep learning have become the development trend in the field of computer vision now and in the future. 
Face recognition combined with deep learning requires high computational power. It is relatively sufficient to develop computational power on a PC, but it is more difficult to deploy on embedded terminals . It requires comprehensive consideration of real-time performance，accuracy， computational power，power consumption，cost, portability, development difficulty and other factors. 
This paper aims at the above problems, designs and implements a PYNQ-based face recognition system, which realizes face recognition combined with deep learning in embedded terminals—real-time video input, face recognition, and display result output. 
The main tasks of the project are: 1. Investigate the background, significance, and application prospects of face recognition, and determine the goals of the project; 2. Analyze the characteristics of traditional face recognition algorithms and  algorithms combined with deep learning.Implement them on the PC side, and verify and contrast the results. 3. Propose a design and implementation scheme of PYNQ-based embedded terminal face recognition system, and analyze the design and implementation principle of each part of the system in detail; 4. Set up an embedded terminal face recognition system to test and verify the project and analysis the results 5. A summary of the issues and outlook, proposed system improvement and optimization methods. 
The scheme performs well in real time, accuracy, power consumption and cost. In the future, face detection and recognition in algorithms can be tried to use YOLO, SSD ,BNN ,Resnet and so on, and the real-time and accuracy will be better. If the cost and power consumption requirements of the development platform are not strict, Nvidia's GPU embedded development platform—Jetson, should be tried, and the performance should be improved greatly and the development will be easier. Of course, only using FPGA, the development is difficult, but the performance meets the requirements and the cost and power consumption is lower, we should try to break through. 
This project belongs to the embedded realization of computer vision algorithm combined with deep learning, This will be followed by more in-depth research and implementation based on this project in this direction.