Shadow Profiles
Shadow profiles are automatically created for people detected by cameras who are not yet linked to a known member. The Shadow Profiles settings page controls how selective the system is when creating these profiles — covering image quality, pose requirements, and linking thresholds.
Preset Configurations
| Preset | Description |
|---|---|
| Lenient | Accepts more visitors as shadow profiles. Lower matching confidence required, wider pose angles allowed, and fewer image quality restrictions. Best for budget cameras, low-light venues, or when maximising profile capture is the priority. |
| Balanced (Recommended) | Recommended for most facilities. Requires clear, front-facing photos with visible facial features. Balances profile capture rate with accuracy — reduces duplicates while still detecting most visitors. |
| Strict | Requires high-confidence face detection, near-frontal pose, and strong image quality. Produces fewer but more accurate and reliable shadow profiles. Best for well-lit entrances with well-positioned cameras. |
Use Reset to Defaults to restore original values. If any threshold has been manually adjusted, the preset label shows Custom.
Self-Learning
| Setting | Default |
|---|---|
| Self-Learning toggle on/off | Enabled |
When enabled, the system automatically adds high-quality face detections as additional photos to existing shadow profiles over time. Only photos that pass all quality and pose checks with 99%+ match confidence are used.
Threshold Settings
Profile Linking Thresholds
When a face is detected and does not match a known member, these thresholds control whether it is linked to an existing shadow profile or treated as a new visitor.
| Setting | Description | Default |
|---|---|---|
| Duplicate Prevention Threshold | Minimum score an existing shadow profile must reach in the final duplicate check. If matched above this, the face links to the existing profile. | 90 |
| Profile Link Threshold | Minimum similarity score to directly link a face to an existing shadow profile without creating a new one. | 80 |
Face Detection Gate
| Setting | Description | Default |
|---|---|---|
| Minimum Detection Confidence | Minimum confidence the AI must have that it has found a real face before any further processing begins. Detections below this are dropped immediately. | 90 |
Pose Requirements
Maximum pose angles allowed for a face to be accepted for shadow profile creation. These limits should generally be stricter than Member Matching, as they control the initial reference photo used to create the profile.
Note: If too lenient, poor-angle shots will be used as reference photos, which can cause different people to be grouped under the same shadow profile.
Quality Checks
Two-stage image quality assessment runs only when no existing shadow profile match is found. Both stages must pass before a new shadow profile can be created.
Tier 1 — Fast Checks
Quick checks that run in milliseconds to reject obviously unusable photos before any further processing.
| Setting | Description | Default |
|---|---|---|
| Min Resolution (px)Checks the shortest dimension (width or height) of the cropped face image. Tiny face crops from far-away cameras contain too little detail for reliable quality assessment. Lower = more permissive; higher = only accepts larger, closer face crops. | Minimum pixel width or height the face crop must have | 80 |
| Min SharpnessThe system runs an edge-detection algorithm on the image. Blurry or out-of-focus photos have very few sharp edges and score low. A higher threshold means only crisp, well-focused photos get through. | How in-focus the photo must be | 5 |
| Min ContrastMeasures how much difference there is between the lightest and darkest areas of the image. A flat, washed-out photo — like a face under dim, even grey lighting — scores low. A higher threshold means the image needs a meaningful spread of tones to pass. | How much variation in light and dark the image must have | 18 |
| Min Lighting SymmetryCompares the average brightness of the left half of the image against the right half. If one side of the face is dramatically brighter than the other — for example, lit hard from a single side — this score drops. The value is 0 to 1, where 1 means perfectly balanced. Higher = stricter about even left-to-right lighting. | How evenly lit the left and right sides are (0–1) | 0.55 |
| Max UnderexposedMeasures what fraction of the image is essentially pitch black. For example, a value of 0.38 means up to 38% of pixels can be near-black before the photo is rejected. Lower = stricter — fewer dark or underexposed photos will make it through. | Maximum proportion of near-black pixels (0–1) | 0.45 |
| Max OverexposedThe same concept as underexposure, but for near-white or blown-out pixels. For example, a value of 0.28 means up to 28% of pixels can be completely washed out before the photo is rejected. Lower = stricter — fewer overexposed photos will make it through. | Maximum proportion of near-white pixels (0–1) | 0.38 |
| Min Dynamic RangeChecks whether the image uses a wide spread of the brightness spectrum, from dark areas to bright areas. If the whole image sits in a narrow band — for example, everything is a similar mid-grey shade — this check fails. Higher = requires more tonal variety across the image. | How broadly the image uses the full tonal range (0–1) | 0.3 |
| Min Eye Aspect RatioMeasures how open the eyes are by comparing the height of the eye opening to its width. Closed or squinting eyes produce a very low ratio. A higher threshold requires the eyes to be clearly open and visible. | How open the eyes must be | 0.1 |
Tier 2 — ISO Quality Assessment
In-depth face quality checks based on the ISO biometric standard. Each value is a quality score from 0–100 representing how well the photo meets that criterion. Only runs on photos that pass Tier 1.
| Setting | Description | Default |
|---|---|---|
| Face OcclusionAn ISO quality score — not a degree or percentage. 100 means the face is fully visible with nothing blocking it; lower scores indicate increasing obstruction. The value you set is the minimum score a photo must achieve. Checks for sunglasses, face masks, hands, scarves, or anything else covering part of the face. | How unobstructed the face is (higher = rejects masked/partially hidden faces) | 45 |
| Eyes VisibleAn ISO quality score — not a degree or percentage. 100 means both eyes are fully visible and unobstructed; lower scores indicate the eyes are partially or fully obscured. Useful for catching sunglasses, hair across the eyes, or eyes that are partially out of frame. | How clearly the eyes are visible | 40 |
| Mouth OcclusionAn ISO quality score — not a degree or percentage. 100 means the mouth is fully visible; lower scores indicate the lower face is covered by a mask, hand, scarf, or other object. | How visible the mouth area is | 30 |
| Head Pose YawAn ISO quality score representing how front-facing the head is left-to-right — not an angle in degrees. 100 means looking perfectly straight at the camera; lower scores mean the head is increasingly turned to one side. The value you set is the minimum score required, not a maximum angle. Compare this to the "Face Pose Limits — Identified" settings above which are actual degree thresholds. | How directly the person faces the camera left-to-right | 60 |
| Head Pose PitchAn ISO quality score representing how level the head is tilted up or down — not an angle in degrees. 100 means looking perfectly straight ahead; lower scores mean the head is increasingly tilted up or down. Useful for ceiling-mounted cameras where people naturally look upward. Compare this to the "Face Pose Limits — Identified" settings which are actual degree thresholds. | How level the head is up and down | 60 |
| Head Pose RollAn ISO quality score representing how upright the head is — not an angle in degrees. 100 means the head is perfectly level with no sideways tilt; lower scores mean the head is increasingly tilted to one side. Compare this to the "Face Pose Limits — Identified" settings which are actual degree thresholds. | How upright the head is | 80 |
| Illumination UniformityAn ISO quality score for how evenly the face is lit. Higher values reject faces with harsh shadows, strong side-lighting, or uneven illumination across the face. | How evenly the face is lit | 40 |
| Over-Exposure PreventionAn ISO quality score for how free the face is from blown-out highlights. Higher values reject photos where parts of the face are washed out due to strong backlighting or direct flash. | How free the face is from blown-out highlights | 35 |
| Compression ArtifactsAn ISO quality score for how free the image is from compression distortion. Higher values reject blocky, pixelated, or heavily compressed images — common with low-bitrate camera streams. | How free the image is from compression distortion | 50 |
| Expression NeutralityAn ISO quality score for how neutral the facial expression is. Higher values reject photos with strong expressions like smiling, frowning, or raised eyebrows, which can alter facial geometry and reduce matching accuracy. | How neutral the facial expression is | 5 |
| Minimum Overall QualityOFIQ's holistic biometric quality score (0–100). Rejects images that are fundamentally unusable for face recognition — blurry, too small, or too poorly lit. Calibrated for security cameras, not passport photos. | OFIQ holistic biometric quality score (0–100) | 15 |
Reject IR / Night Vision Photos — when enabled, images captured in infrared or night vision mode are rejected. Disable this if your cameras use IR mode and you still want shadow profiles created from those images.
Logic Visualisation
A diagram on the right side of the page shows the full shadow profile decision flow with your current settings applied — from face detection through each gate to either a new shadow profile or a link to an existing one. Each decision point links back to the relevant setting.