Skip to main content

Face & Mask Detection

The Face & Mask Detection action provides two related capabilities: detecting human faces in video streams and checking for compliance with safety gear requirements (masks, headwear).

Overview​

This action is essential for:

  • Security applications requiring face detection
  • Health and safety compliance (mask wearing)
  • PPE (Personal Protective Equipment) verification in work environments

The action includes two separate functions that can be used independently or together in a workflow.

Feature 1: Face Detection​

What It Does​

The Face Detection function scans video frames to locate human faces. It's optimized for:

  • Multiple faces in a single frame
  • Various face orientations (though frontal faces work best)
  • Different lighting conditions

Configuration Options​

ParameterTypeDefaultDescription
StreamsTable SelectionRequiredCamera(s) to analyze
Confidence ThresholdNumber (0-1)0.5Minimum confidence required to report a face
Max Runtime SecondsNumber60How long to search before giving up

Results​

ResultMeaning
face_detectedAt least one face was found in the video stream
no_face_detectedNo faces were found during the entire runtime

Use Cases​

  • Access Control: Detect when someone approaches a door
  • Attention Detection: Verify someone is present at a station
  • Pre-processing: Capture face images for further analysis

Feature 2: Head/Work Clothes Detection​

What It Does​

The Head/Work Clothes Detection function analyzes a previously captured image to check for specific headwear or masks. This includes:

  • Medical/surgical masks
  • N95 respirators
  • Hard hats
  • Hair nets
  • Other head-mounted PPE

Important Prerequisite​

This function must run after a detection action that captures an image. It cannot analyze live video directly—it needs an image from a previous action in the workflow (such as Face Detection or Object Detection).

Configuration Options​

ParameterTypeDefaultDescription
Confidence ThresholdNumber (0-1)0.5Minimum confidence required to report detection

Results​

ResultMeaning
head work clothes detectedMask or headwear was found in the image
no head work clothes detectedNo mask or headwear was detected

Typical Workflow​

A complete PPE compliance workflow might look like this:

  1. Object Detection: Detect a person entering the work area
  2. Face Detection: Confirm a face is visible (person facing camera)
  3. Mask Detection: Check if the person is wearing required PPE
  4. Decision Branch:
    • If mask detected → Log compliance and end
    • If no mask → Create Violation and alert supervisor

Common Use Cases​

Health Compliance (Medical/Food Service)​

  • Scenario: Ensure staff wear masks in food preparation areas
  • Setup: Camera at kitchen entrance, face detection zone
  • Workflow: Detect person → Check for mask → Alert if non-compliant

Construction Safety​

  • Scenario: Verify hard hat usage on construction sites
  • Setup: Camera at site entry point
  • Workflow: Detect person → Check for hard hat → Log or alert

Cleanroom Compliance​

  • Scenario: Ensure proper head covering in controlled environments
  • Setup: Camera at cleanroom entrance
  • Workflow: Detect person → Check for hair net/cap → Grant or deny access alert

Troubleshooting​

Low Detection Accuracy​

  1. Lighting Issues: Ensure faces are well-lit from the front. Backlighting (bright light behind the person) severely impacts accuracy.

  2. Camera Angle: Face detection works best when faces are relatively frontal. Extreme angles (top-down or side views) reduce accuracy.

  3. Distance: Faces that are too small in the frame (person far from camera) may not be detected. Adjust camera zoom or position.

  4. Obstructions: Sunglasses, large hats, or hair covering the face can interfere with detection.

Mask Detection Not Working​

  1. Check Workflow Order: Mask detection requires a previous action that provides an image. Ensure Face Detection or Object Detection runs first.

  2. Image Quality: Blurry or low-resolution images make mask detection difficult.

  3. Mask Types: The system is trained on common mask styles. Unusual designs may not be recognized.

  4. Partial Visibility: If the mask is only partially visible (person turning away), detection may fail.

Model Loading Errors​

  1. Check Model Files: Verify that the required model files are present on the server:

    • face_detection.pt for face detection
    • mask_detection_model.h5 for mask detection

  2. Memory Issues: Large AI models require sufficient RAM. Check server memory availability.

  3. File Permissions: Ensure the ResEngine process has read access to model files.

Too Many False Positives​

  1. Increase Confidence Threshold: Raise the threshold to 0.6 or 0.7 to be more selective.

  2. Improve Lighting: Inconsistent lighting can cause shadows that resemble faces.

  3. Check for Face-Like Patterns: Posters, photos, or patterns that resemble faces can trigger false detections. Adjust zones to exclude these areas.

Missing Detections​

  1. Lower Confidence Threshold: Try reducing to 0.3 or 0.4 to catch more detections.

  2. Increase Runtime: Allow more time for detection by increasing Max Runtime Seconds.

  3. Check Camera Coverage: Ensure people are facing the camera when they enter the detection zone.

Best Practices​

  1. Camera Placement: Position cameras at face height (approximately 1.5-1.8 meters) for optimal face capture.

  2. Consistent Lighting: Use even, front-facing lighting. Avoid backlighting and harsh shadows.

  3. Clear Signage: Inform people that face/mask detection is in use for transparency and compliance.

  4. Regular Testing: Periodically verify the system is detecting correctly, especially after camera adjustments or lighting changes.

  5. Combine with Object Detection: For better reliability, use Object Detection to first confirm a person is present, then run Face Detection on the cropped person image.