Fine-Tuning and Troubleshooting Vivotek Facial Recognition
Vivotek's facial recognition technology is a powerful tool for enhancing security, providing identity verification, and enabling smart access control. However, achieving high accuracy and reliability requires careful setup and configuration. If your system is failing to recognise faces, generating false alerts, or performing poorly, this guide will help you troubleshoot and optimise its performance.
Effective facial recognition depends on a perfect synergy between camera placement, lighting, software configuration, and the quality of your face database. A weakness in any one of these areas can significantly degrade the system's effectiveness.
### Common Facial Recognition Issues
- Low Accuracy: The system frequently fails to match a person to their profile in the database.
- False Positives: The system incorrectly identifies a person or even an object as a registered individual.
- Missed Detections: People walk past the camera, and the system fails to detect a face at all.
- Environmental Challenges: Performance drops significantly in low light, bright backlight, or at sharp angles.
- Database Errors: Problems with uploading, managing, or syncing the facial database.
Part 1: Optimising Camera Installation and Environment
The foundation of accurate facial recognition is a high-quality image. The camera's physical setup is paramount.
## 1. Camera Placement and Angle
- Height: Mount the camera at a height that provides a frontal view of a person's face. The ideal height is typically between 2.5 and 3.5 metres (8-11.5 feet), depending on the lens and the expected distance of the subject.
- Angle: The camera should be angled as close to level with the subjects' faces as possible. Avoid steep downward angles (like from a high ceiling), as this will capture the top of heads rather than faces. The vertical angle should be less than 20 degrees.
- Facial Pixels: For reliable recognition, the face of the subject should have a resolution of at least 80x80 pixels. Position the camera close enough to the target area to achieve this.
## 2. Lighting is Critical
- Avoid Backlight: Do not point the camera directly at a strong light source, such as the sun or bright indoor lights. Strong backlight will create silhouettes and make facial features impossible to distinguish.
- Ensure Even Illumination: The target area should be evenly and brightly lit. Use diffused, consistent lighting to avoid harsh shadows on faces, which can obscure key features. Vivotek's SNV (Supreme Night Visibility) and WDR (Wide Dynamic Range) technologies can help, but they cannot overcome extremely poor lighting.
- Use IR with Caution: While IR (infrared) can help detect faces in the dark, recognition accuracy may be reduced compared to well-lit visible light conditions.
Part 2: Software Configuration and Database Management
Once the camera is physically optimised, you need to fine-tune the software settings in Vivotek's VMS (Video Management Software), such as VAST Face or Shepherd.
## 1. Building a High-Quality Face Database
The system is only as good as the data it has to work with. A poorly constructed database is a primary cause of low accuracy.
- Use High-Quality Photos: The reference images you upload to the database should be clear, high-resolution, and well-lit.
- Frontal Pose: Use passport-style photos where the person is looking directly at the camera.
- Neutral Expression: The person should have a neutral facial expression.
- No Obstructions: Avoid images where the person is wearing sunglasses, a hat, a mask, or has hair covering their face.
- Multiple Images: If possible, add more than one image of each person, perhaps with slightly different lighting or expressions, to improve matching reliability.
## 2. Calibrating Recognition Settings
- Confidence Threshold: In the software settings, you can usually adjust the confidence threshold for a match. A lower threshold will result in more matches (and potentially more false positives), while a higher threshold will be more selective but may miss some correct matches. Adjust this based on your specific security needs.
- Detection Zones: Configure the facial recognition analytics to only run within a specific area of the camera's view. This prevents the system from wasting processing power on irrelevant parts of the scene and can reduce false detections from posters or reflections.
## 3. Setting Up Events and Alarms
If you are not receiving alerts, the issue is likely in your event configuration.
- Create a Rule: In your VMS, you must create a specific rule for facial recognition events.
- Define the Trigger: Set the trigger as 'Facial Recognition Match' (for recognised individuals) or 'Unrecognised Face Detected'. You can often link this to specific groups (e.g., 'Staff', 'VIP', 'Blocked List').
- Define the Action: Configure what happens when the event is triggered. This could be a push notification, an email alert, a bookmark in the recording, or triggering a relay for access control. Ensure this action is correctly set up and enabled.
## 4. System Maintenance
- Update Firmware and Software: Regularly check for and install the latest firmware for your Vivotek camera and updates for your VMS software. These updates often include significant improvements to the analytics engines.
- Clean the Lens: A smudged or dirty camera lens will degrade image quality and impact recognition accuracy. Clean it regularly with a microfibre cloth.