With the rapid development of artificial intelligence, the research and application of computer vision technology have gradually entered the mature stage. Among them, face recognition is a more widely used technology, has penetrated into all aspects of human daily life. TSINGSEE Qingxi Video is also actively developing face recognition projects, integrating face recognition technology into related video platforms (such as EasyCVR video fusion cloud service) and applying it to offline scenes.
This article will briefly share with you: how is face recognition done? What is its flow?
First, the composition of face recognition system
- Front-end image collection Front-end face image collection system collects face images and real-time video streams by capturing, such as video surveillance cameras, intelligent attendance machines, intelligent access control machines and other devices, all have the function of face capture.
- Back-end intelligence platform back-end intelligence platform related data from the front end can be unified for gathering, processing, storage, use, management and sharing, and connecting with the face recognition system, realize face recognition function, is applied in real scenarios such as face recognition, face of check on work attendance access control, the Ministry of Public Security of the human face tracking, capturing the suspect, and so on.
Two, what is the process of face recognition technology?
1, face detection
In practice, face detection is mainly used for the pretreatment of face recognition, that is, to accurately calibrate the location and size of the face in the image. From the photo to find the location of the face, with the picture of the upper left corner as the origin of coordinates, respectively recorded the coordinates of the upper left corner and the lower right corner of the face frame, and will be part of the face cropped out.
2, face alignment
In real scenes, the Angle of the face captured by the front-end device is not necessarily the straight face, so the face posture in the image needs to be corrected. Through the key point detection of the face of the key point coordinates, according to the key point coordinates of the face to adjust the Angle of the face, so that the face alignment. In this picture, these two faces, from the computer’s point of view, are completely different faces, so we need to do some affine transformation to align the faces.
- 1) affine transformation function of affine transformation is from 2 d coordinates to linear transformation between 2 d coordinates, and keep the 2 d graphics “laymen ping” and “parallel” (line remains the same, the relative position of the relationship between parallel lines after affine transformation is still parallel lines, and the position of the line on the order will not change).
- 2) The alignment method uses the trained model to automatically mark 68 landmarks from the detected faces. Then, a standard template is searched in the template library, and affine changes are used to align the 68 landmarks with the 68 landmarks of the template.
3, face coding (feature vector extraction)
The convolutional neural network is used to train a model, and the face image is automatically encoded into a 128 dimensional vector with a strong semantic meaning. Training methods:
- Enter a photo of a known identity.
- Enter a photo of the same identity.
- Enter a photo of a different identity.
- Adjust the parameters repeatedly so that the photo codes in Step 1 and Step 2 are as close to each other as possible and different from those in Step 3.
4, face classification
Compute the Euclidean distance of the 128 bit vector difference of each image in the input image and database in turn, until find the one less than our threshold value, to this, face recognition is successful.
5. Experimental results
Based on the above steps, we will detect the final recognition results of the face recognition system.
Three, the application scenario of face recognition technology
Face recognition technology is mainly used for identity verification, common scenes are residential areas, buildings, campuses, factories, parks, banks and so on, such as: intelligent access control, face gate, face attendance, intelligent door locks and so on. Through face recognition to verify identity, to ensure the safety of relevant places, but also reduce the cost of manual audit.
In security monitoring, face recognition is also significant, such as public places (subway stations, stations, streets, hotels, etc.) security control, the Ministry of Public Security chase suspects. Based on security surveillance cameras in public places, it can capture faces and upload the results to the network of the Ministry of Public Security for comparison with the faces of suspects to assist the law enforcement work of public security personnel.
Face recognition technology is a key technology in the field of artificial intelligence and has a very wide application prospect in intelligent video surveillance system. TSINGSEE Video will also continue to develop intelligent business systems and platforms in multiple scenarios, focusing on AI detection and recognition technology.