Design And Implementation Of Recognition System For 3D With Partial Occlusion

In recent years,3D facial recognition technology has become a research hotspot in the field of facial recognition technology.Compared to 2D facial recognition technology,3D facial recognition has higher recognition accuracy and robustness,and has been widely applied in fields such as security,human-computer interaction,virtual reality,and so on.However,there are still some challenges at present.In partially occluded scenes,some features of the face are lost,which affects the recognition accuracy of 3D face recognition technology.In addition,3D face recognition algorithms require many computational resources,so it is necessary to consider the issue of computational overhead when applying them on a large scale.To accurately identify masks occluding faces in actual scenes,this thesis uses 3D reconstruction technology to construct a 100000-level mask occluded 3D face database.The 3D face database is used as a training set to implement a locally occluded 3D face recognition method.This method introduces feature fusion technology,which can effectively combine facial geometry features and depth features to improve the accuracy of face recognition in locally occluded scenes.In addition,this thesis proposes a lightweight model-based face recognition frame filtering mechanism,which consists of background frame filters,face filters,and duplicate person filters.The face filter is an improved face detection model based on SSD network,which uses bicubic interpolation to improve the accuracy of feature maps and fuses some feature maps to facilitate low-quality face detection,this mechanism reduces the video frame processing load of the 3D face recognition model by filtering out irrelevant and redundant video frames,saving server storage and computing costs.Finally,this thesis designs and implements a recognition system for 3D face with partial occlusion based on the B/S architecture.The system integrates local occluded 3D facial recognition algorithms and frame filtering mechanisms,and its functions are designed in a modular manner,including real-time video monitoring,device management,personnel management,facial database management,institutional management,label management,history recording,and user management modules,implemented complete facial recognition data management function and provided a friendly user interface.This thesis tested the performance of the partially occluded 3D face recognition algorithm and frame filtering mechanism through experiments and verified the overall functionality of the system.The experimental results showed that the partially occluded 3D face recognition algorithm has high accuracy and robustness in processing occluded faces,and the system can handle more video streams of data in resource limited scenarios.