Face Search and Recognition Solution

Face Search enables the rapid search of a person in a video footage based on an input image of the person’s face. With the capability of the search engine based on artificial intelligence and deep learning, each search query can yield results within only seconds, which helps locating a person almost immediately in real time. AI-powered Face Search avoids privacy concerns by not storing any faces in the search database, i.e., there is no list of pre-stored faces to compare. The Face Search algorithm compares the target face against all faces in the recorded video and not against a known face database. Names are not stored to avoid privacy regulations and restrictions.
Face Search is a simple and lightweight solution for shopping centers, schools, resorts, airports, etc. to increase business efficiency and minimizes manual effort in searching for people when incidents occur on site.
The input image can be uploaded from an external source (e.g., images saved on phones) or a snapshot from the surveillance video feed that contains the face of interest. The image should contain a full frontal view of the face of interest. It can contain multiple persons and faces, and the Face Search UI allows the user to select the face of interest among all detected faces in the image.

Acceptable formats: all common image formats (e.g., .JPEG, .JPG, .PNG)
Size: the face in the upload image and video should be at least 120 px for high accuracy (> 95%).

The input image can also be obtained from the camera stream, such as when the security staff knows that a person has appeared in the building, would like to search for her from 11 am to midnight, but he does not have an image of her face to upload. In this case, AI-powered Video Search, which enables the detection of general faces, provides the image, which is used for facial feature analysis in AI Face Search to find similar appearances in the recorded video. Thus, no image needs to be uploaded or downloaded.

AI-Powered Face Search Advantages:
1. High efficiency: a target person can be located within a few seconds, which enables fast response time;
2. High performance: high accuracy in a large database; stable performance
3. Easy setup and usage: Smart video search appliance with built-in face search engine can be customized to integrate to any existing NVR/VMS/camera system or as a standalone unit depending on the customer’s needs. The setup can be completed within the hour. The simple-to-use interface requires minimal training and no special programming skills.
4. High cost saving: one-time cost for an AI-powered video search appliance; minimal cost for manual effort to operate; minimal cost for system integration and maintenance.
5. Privacy: Face Search is not face recognition. Because Face Search is not based on face recognition; no faces and name identifications are stored, so Face Search can be used in public environments to identify faces against past video recordings.

Capability of Face Search Engine:
 

1. Continuously processes all faces recorded by the cameras and streamed into the system and maintains the records in the database.
2. Help the user search for certain faces in 2 ways:

• by uploading an image of a person of interest
• by selecting a detected face on camera. In both cases, the Face Search system checks all selected history records (e.g., up to 30 days) and returns the face images that match the target person.

In Face Recognition, when a face appears on camera, the AI will directly match the detected face to the faces in the face database and/or black/white lists to determine whether an alert (if configured) is triggered.

Face Recognition Solution enables the simultaneous identification of multiple people of interest in real time across several live camera streams. AI-based Face Recognition System continuously detect and recognize faces of the detected people from IP video surveillance cameras, at a distance, on the move in natural settings. High performance face recognition software using deep learning technology can identify people across multiple cameras in < 2 seconds to enable fast response time and provide higher than 90 percent accuracy for environment with good camera angles and lighting.

Face Recognition Solution Advantage:
 

• Easy to implement Face Recognition Function to your current CCTV system.
• Support multiple channel cameras share one Face Recognition software running on the platform to save license costs.
• With Intelligent video filter, only images with human faces are fed to the Face Recognition engine to maximize efficiency.
• Easy to integrate with any ONVIF standard based IP cameras

Face Recognition Performance
 

• Face match rate high than 90% with good camera angles and lighting
• Minimum processing performance: 10 fps for multiple video streams and multiple faces per image.
• Maximum processing performance: 500 fps on a Video Search AI Box (for 8 channels) with multiple GPUs in one server and HD resolution video stream.
• Live face extraction in 1 millisecond per face and comparison of the extract face to a 50,000-face database in 1 second using one server.
• Forensic (post-process) video face recognition: 8x faster than live video processing.

Biometric Facial Recognition characteristics:
Biometric identification through a video camera commonly faces a variety of challenges because the face images captured by the camera system are naturally prone to variability with regard to the reference images enrolled in the reference photo library.
AI-powered Face Recognition software algorithm can handle the following sources of image variation:
 

• Changes in facial expression
• Changes in beard and hairstyle
• Changes due to moderate aging
• Partial occlusions of the face, such as hats, scarves, caps, glasses and sunglasses
• Variations in lighting conditions and ambient illumination down to 1lux
• Head pose deviations up to ±35º horizontally and vertically
• Blur (out of focus or motion)
• Video artifacts due to compression

Require Image Quality:
To have the best recognition result, the reference images in the face recognition solution should have high quality and conform to the prescribed standard. Color information is not a requirement for face recognition, so grayscale images of at least 8 bits (256 gray values) are supported. Since the face sizes may vary from one person to another, the face resolutions are typically estimated by measuring the distance between the centers of the eyes (interocular distance) in pixels.
The camera angle and focal length also affect the face recognition result. In order to obtain better recognition result, the pixel size of the face should be up to 120 pixel. The following figure show the recommended values for setup to have the best performance.

Face Recognition System Functionalities
AI-powered Face Recognition System Functionalities