Face recognition system

Abstract

The invention provides a face recognition system, and relates to the technical field of face recognition, and the face recognition system comprises the following face detection steps: step 1, image feature extraction; step 2, face positioning; step 3, key point detection; step 4, face alignment is carried out; step 5, face feature extraction; step 6, similarity calculation; and step 7, virtual reality combination. The method has the advantages that the RetinaFace can accurately detect a human face in a complex environment through the efficient multi-task learning ability, the ArcFace provides high-precision human face feature extraction and is suitable for human face recognition tasks of various scenes, and the EfficientNet performs human face expression recognition based on human face detection and human face recognition results, so that the human face recognition efficiency is improved, and the human face recognition efficiency is improved. And the age and the gender of the figure are predicted, so that the accuracy rate of people in a complex scene with a large flow of people is up to 99%.

Description

Technical Field

[0001] The present invention relates to the technical field of face recognition, and particularly to a face recognition system. Background Art

[0002] A face recognition system is a biometric recognition technology system that performs identity recognition based on the facial feature information of a person. By using a camera or webcam to collect images or video streams containing human faces, it automatically detects and tracks faces in the images, and then performs a series of processing and analysis on the detected faces to achieve identity confirmation or search.

[0003] Most existing face recognition systems can only be used when there are few people. Because when there are many people, the system will be relatively complex and chaotic, which will affect the capture effect of the system, resulting in the system being prone to fusing and recognizing the facial features of multiple people during the recognition process, thus leading to errors in the system's recognition and affecting the recognition effect. Summary of the Invention

[0004] The purpose of the present invention is to solve the drawbacks existing in the prior art, and a face recognition system is proposed. The technical solution adopted by the present invention is as follows:

[0005] A face recognition system, the steps of face detection include:

[0006] Step 1: Image feature extraction;

[0007] Step 2: Face localization;

[0008] Step 3: Key point detection;

[0009] Step 4: Face alignment;

[0010] Step 5: Face feature extraction;

[0011] Step 6: Similarity calculation;

[0012] Step 7: Virtual reality combination.

[0013] As an improvement, the specific steps of the said Step 1: Image feature extraction are: extracting the features in the image through a deep convolutional neural network.

[0014] As an improvement, the specific steps of the said Step 2: Face localization are: generating candidate boxes according to the features, and applying the non-maximum suppression algorithm to remove redundant boxes to accurately locate the face area.

[0015] As an improvement, the specific steps of Step 3: Key Point Detection are as follows: In addition to the face bounding box, RetinaFace also detects the key points of the face to support subsequent face alignment. When returning the key points, we will add corresponding special effects to the key points to make the presented video effect.

[0016] As an improvement, the specific steps of Step 4: Face Alignment are as follows: According to the detected key points, the face is normalized to ensure the consistency of the input data.

[0017] As an improvement, the specific steps of Step 5: Face Feature Extraction are as follows: The face feature vector is extracted through the ArcFace model to distinguish different individuals. After extracting the features of each face, a unique face_id is assigned to each face for subsequent recognition and comparison.

[0018] As an improvement, the specific steps of Step 6: Similarity Calculation are as follows: The extracted face feature vector is compared with the existing face features in the database, the cosine distance is calculated, and the cosine distances are sorted from smallest to largest. The smaller the distance, the more similar. Finally, the top 5 faces in the sorting are returned as the final result.

[0019] As an improvement, the specific steps of Step 7: Virtual Reality Integration are as follows: Using the interface provided by the AI Doubao model, we specifically combine different scenarios to add virtual backgrounds corresponding to the scenarios and generate virtual images that match the facial features of the characters. The combination of virtual and real enhances the tourist experience. At the same time, when the tourist's visit ends, we will generate the tourist's visit footprint and customize a visit strategy for the tourist, and generate a virtual tour guide for explanation, explaining the historical and cultural background of the scenic spot, local customs, food recommendations, etc. Different from traditional voice-guided tour devices, the AI tour guide can not only interact according to real-time questions but also provide more flexible and personalized answers.

[0020] The beneficial effects of the present invention are as follows:

[0021] A face recognition system of the present invention. RetinaFace, with its efficient multi-task learning ability, can accurately detect faces in complex environments, while ArcFace provides high-precision face feature extraction and is suitable for face recognition tasks in various scenarios. EfficientNet performs facial expression recognition of people based on the results of face detection and face recognition, and then predicts the age and gender of the people. The combination of these three makes the accuracy rate reach 99% in complex and crowded scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flowchart for face_id assignment of the present invention.

[0023] Figure 2 This is the face processing flowchart of the present invention. Detailed implementation manners

[0024] In order to make the content of the present invention easier to be clearly understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. The same components are denoted by the same reference numerals. It should be noted that the words "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to the directions in the drawings, and the words "inner" and "outer" respectively refer to the directions towards or away from the geometric center of a specific component.

[0025] As Figure 1 shown:

[0026] The steps of face detection using RetinaFace are as follows:

[0027] Feature extraction: Extract features in the image through a deep convolutional neural network.

[0028] Face localization: Generate candidate boxes according to the features, and apply the non-maximum suppression (NMS) algorithm to remove redundant boxes to accurately locate the face area.

[0029] Key point detection: In addition to the face box, RetinaFace also detects the key points of the face (such as the positions of eyes, nose, and mouth), provides support for subsequent face alignment, and adds corresponding special effects to the key points while returning the key points to make the presented video effect.

[0030] After detecting the face, use ArcFace for face recognition:

[0031] Face alignment: Standardize the face according to the detected key points to ensure the consistency of the input data.

[0032] Feature extraction: Extract the face feature vectors through the ArcFace model to distinguish different individuals. After extracting the features of each face, assign a unique face_id to each face for subsequent recognition and comparison.

[0033] As Figure 2 shown:

[0034] Similarity calculation: Compare the extracted face feature vectors with the existing face features in the database, calculate the cosine distance, and sort the cosine distances from small to large. The smaller the distance, the more similar. Finally, return the top 5 faces before sorting as the final result.

[0035] Others: By using the interface provided by the Doubao large model, we specifically combine different scenarios to add virtual backgrounds corresponding to the scenarios and generate virtual avatars that match the facial features of the people. The combination of virtual and real enhances the tourists' experience. At the end of the tourists' visit, we will generate a visit footprint for them, customize a visit guide, and generate a virtual tour guide to explain the historical and cultural background of the scenic spots, local customs, food recommendations, etc. Different from traditional voice-guided tour devices, the AI tour guide can not only interact based on real-time questions but also provide more flexible and personalized answers.

[0036] RetinaFace, with its efficient multi-task learning ability, can accurately detect human faces in complex environments, while ArcFace provides high-precision face feature extraction and is suitable for face recognition tasks in various scenarios. EfficientNet performs facial expression recognition of people based on the results of face detection and face recognition, and then predicts the age and gender of the people. The combination of these three enables us to achieve an accuracy rate of up to 99% in complex and crowded scenarios.

[0037] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A face recognition system, characterized in that: The steps of face detection include: Step 1: Image feature extraction; Step 2: Face positioning; Step 3: key point detection; Step 4: Face alignment; Step 5: facial feature extraction; Step 6: Similarity calculation; Step 7: Virtual reality integration.

2. A face recognition system according to claim 1, characterized in that: The specific steps of step 1: image feature extraction are: extracting features in the image through a deep convolutional neural network.

3. A face recognition system according to claim 1, characterized in that: The specific steps of step 2: face positioning are: generating candidate frames according to features, and applying a non-maximum suppression algorithm to remove redundant frames to accurately position the face area.

4. A face recognition system according to claim 1, characterized in that: The specific steps of step three: key point detection are: in addition to the face frame, RetinaFace also detects the key points of the face to provide support for subsequent face alignment. When returning the key points, we will add corresponding special effects to the key points to improve the presented video effect.

5. A face recognition system according to claim 1, characterized in that: The specific steps of step 4: face alignment are: based on the detected key points, the face is standardized to ensure the consistency of the input data.

6. A face recognition system according to claim 1, characterized in that: The specific steps of step 5: facial feature extraction are: extracting facial feature vectors through the ArcFace model to distinguish different individuals, and after extracting the features of each face, assigning a unique face_id to each face for subsequent recognition and comparison.

7. A face recognition system according to claim 1, characterized in that: The specific steps of step six: similarity calculation are: comparing the extracted facial feature vector with the facial features already in the database, calculating the cosine distance, and sorting the cosine distances from small to large, where the smaller the distance, the more similar it is, and finally returning the top 5 faces as the final result.

8. The face recognition system according to claim 1, characterized in that: The specific steps of step seven: virtual reality integration are: using the interface provided by the ai bean bag model, we specifically combine different scenes to add virtual backgrounds corresponding to the scenes, and generate virtual images that match the facial features of the characters. The combination of virtual and reality enhances the tourists' experience. At the same time, when the tourists finish their tour, we will generate travel footprints and tailor-made travel guides for them, and generate virtual tour guides to explain the historical and cultural background of the attractions, local customs, food recommendations, etc. Unlike traditional voice guide equipment, AI tour guides can not only interact based on real-time questions, but also provide more flexible and personalized answers.

Images