Research And Implementation Of Web Based Facial Keypoints Real Time Detection System

Real-time facial keypoints detection technology has broad application prospects in the fields of media and entertainment such as augmented reality and human-computer interaction.However,there are still many problems in practical applications.It is difficult for APP to meet the cross-platform requirements of facial detection services,and the delay caused by cloud computing makes it difficult for facial detection meet the real-time requirements.End computing,as an important computing component in the terminal intelligence,becomes a beneficial supplement to cloud computing.Therefore,web-based real-time detection in end computing mode provides a cross-platform,universal solution for facial feature points detection.However,it is the limited computing power of the browser that makes it difficult to run a neural network model in the Web environment.The mainstream object detection neural network model is difficult to meet the real-time requirement in the Web environment because of its large size and massive computation.For the real-time facial feature points detection,it is necessary to use a lightweight object detection network structure and reduce the computation and memory usage of the model through effective neural network model acceleration and compression methods.The research of this paper mainly includes two aspects:On one hand,a lightweight facial feature points detection neural network is designed and implemented.On the other hand,based on the implemented lightweight facial keypoints detection network,a web-based facial keypoints real-time detection system in the end-computing mode is implemented.Finally,related experiments were performed to test the usability of the web-based real-time facial keypoints detection system,the effectiveness of the lightweight facial feature points detection network and the effectiveness of the model acceleration methods.