A real-time dynamic face recognition method and system

Abstract

The invention discloses a real-time dynamic face recognition method and system, and the method comprises the steps: carrying out the real-time detection of the face of an image in a video stream, carrying out the alignment and recognition of the face if the face exists, and restarting the detection if the face does not exist; After face recognition is completed, if a known face exists, tracking the face in the video stream, and if the known face does not exist, intercepting a face image for storage and rstarting detection; In the face tracking process, the video stream is detected for the second time at intervals of M frames, the face with the displacement larger than a set threshold value or newly appearing is re-recognized, the face with the displacement smaller than the set threshold value continues to be tracked, and tracking and recognizing of the reduced face are stopped; During face re-identification, if the template is stored in a known face, tracking is continued, and if the known face does not exist, a face image is intercepted and stored, and detection is restarted. According to the method, the face recognition accuracy is guaranteed, and meanwhile the recognition efficiency is improved.

Description

technical field [0001] The invention relates to the technical field of graphics and graphics processing and machine deep learning, in particular to a real-time dynamic face recognition method. Background technique [0002] Face recognition technology is based on human facial features to detect and classify the input face video stream. First find the face in the video, then further extract the feature information of the face according to the location information of the face, and finally compare it with the feature information of the known face database to identify the identity information of the face. [0003] The research on face recognition technology began in the 1960s. The main theory proposed was based on the geometric structure of the face, and it has not been applied in practice. With the rapid development of computer technology and optical imaging technology, it has extended many More effective deep learning models, and gradually enter the commercial application mark...

AI Technical Summary

Problems solved by technology

[0004] 1. The problem of face detection, the deep learning algorithm with high detection accuracy requires a lot of computing time, which cannot achieve real-time performance on low-configuration CPUs

[0005] 2. Recognition time-consuming problem. To detect and recognize each frame of the video stream is not only a large amount of calculation, but also prone to frame loss or freeze, especially the face in the video moves too fast, which affects the actual monitoring face effect

[0006] 3. The problem of recognition errors. For people with very similar facial features, it is easy to make mistakes in face recognition, which seriously affects the application effect of related products.

[0007] 4. The problem of missing recognition. Usually, face recognition is aimed at frontal faces or side faces with a small angle. When the face deviates from a large angle, it cannot be recognized, and this person is a known face database. , leading to missing such a face

[0008] In addition, they all conducted research on a single module of real-time face recognition, but failed to give a complete solution

[0009] To sum up, it can be seen that the current face recognition technology focuses on improving accuracy while ignoring real-time performance. At the same time, a complete real-time dynamic face recognition system or method has not been proposed.

Images