Research On Face Detection And Key Point Location Technology Based On Miniprogram

Real-time face key points detection technology has broad application prospects in video tracking,augmented reality,face recognition and other fields,but it still faces many problems in actual application deployment.The APP-based deployment method needs to be adapted to different mobile phone systems,which is difficult to meet the cross-platform needs of face key points detection services,and the network delay caused by the Web and cloud computing deployment method is difficult to meet the real-time requirements of face key points detection.As one of the important platforms for edge computing in the 5G era,miniprogram provide a cross-platform and universal deployment solution for face key points detection.However,the main problem of running convolutional neural networks in a miniprogram environment is that the computing power of the miniprogram platform is limited,and the application size needs to be less than 2MB to meet the offline deployment requirements.However,mainstream object detection neural networks are large in size and computationally expensive,which can neither meet the deployment requirements of miniprogram nor meet the real-time requirements of face key points detection.Therefore,the deployment of face key points detection applications in a miniprogram environment requires the design of a lightweight neural network structure,and an effective neural network model compression method to reduce the amount of calculation and volume of the model.The research work of this paper mainly includes three aspects:In terms of face detection,a lightweight feature extraction backbone network is designed.At the same time,in order to ensure the detection accuracy of the lightweight network,the multi-scale pyramid network structure is improved and the receptive field enhancement module is designed to enrich the scale information and context information of the feature maps,thereby improving the detection ability of the lightweight object detection network;in terms of the location of face key points,based on the existing network structure,an improved scheme based on the convolution bottleneck structure is proposed.This solution can effectively reduce the amount of parameters and calculations of the face key points detection network.At the same time,the attention mechanism module is designed to strengthen the feature extraction ability of the lightweight network,thereby ensuring the detection accuracy of face key points;In terms of deployment,the difference between miniprogram technology and Web technology is studied,a real-time detection system for face key points based on miniprogram is designed,and the deployment of convolutional neural networks on miniprogram is successfully implemented.Finally,related experiments were carried out to test the usability of the real-time detection system for face key points based on the miniprogram and the effectiveness of the lightweight neural network structure.