Higher visibility
No meaningful low-visibility signal is supported in the current scene, so the framework reports a broader higher-visibility category without assigning a low-visibility cause.
Camera-based visibility awareness
From fixed-camera imagery to localized visibility interpretation.
VisSense Cam combines a fixed outdoor camera, camera-specific optical calibration, recent images from that same camera, and weather context representative of the deployment site. The framework is designed to identify reduced visibility and provide a concise interpretation of the current scene.
Current framework
From camera image to visibility interpretation.
The framework first determines whether the camera scene supports a reduced-visibility condition. Fog or precipitation is interpreted only after low visibility is detected.
STEP 1 · Collect current information
Current camera image + recent images from the same camera +
relative humidity + precipitation context
↓
STEP 2 · Analyze the camera scene
Check image quality and measure visibility-sensitive optical information such as
contrast, edges, gradients, and distant-scene structure. Compare the scene with
that camera's calibrated clear-reference state.
↓
STEP 3 · Combine optical and atmospheric context
Use camera degradation, recent same-camera change, humidity, and precipitation
information to evaluate whether reduced visibility is supported.
↓
Is reduced visibility supported?
YES · Reduced visibility
NO · No significant low-visibility signal
Reduced visibility detected
The camera and atmospheric context support a low-visibility condition. The next step is to interpret the most likely cause.
Determine the likely condition
Likely fog
Low visibility + no reported precipitation + humid conditions +
supporting optical degradation.
Precipitation-related low visibility
Low visibility + precipitation present. The condition is not labeled as fog.
Cause uncertain
Low visibility is supported, but the available weather evidence is mixed,
incomplete, or insufficient for a confident cause interpretation.
No low-visibility hazard detected
The scene does not show a meaningful low-visibility signal. VisSense Cam reports a broader higher-visibility category without assigning a fog or precipitation low-visibility condition.
↓
STEP 4 · Generate the VisSense Cam operational display
Status → Estimated visibility → Condition → Optical trend →
RH / precipitation → UTC update time
Weather context
Use weather information representative of the camera location.
VisSense Cam does not assume that one weather source represents every camera. Weather data are selected for each deployment site, with localized observations prioritized whenever they are available.
1
Use localized observations first
When a camera has a nearby or collocated weather source, VisSense Cam uses those observations for the variables available at that site.
2
Fill missing variables from the nearest METAR
If a localized source is missing a required variable, the framework can use the nearest available reporting METAR for that missing weather context when camera coordinates are configured.
3
Use METAR when no local weather source exists
For a new camera without localized weather observations, the deployment coordinates can be used to select the nearest reporting METAR as the atmospheric-context fallback.
Weather-source priority:
Localized source
→
Nearest reporting METAR for missing data
→
Unavailable rather than invented
Partial local weather coverage
The sources can be mixed by variable. For example, humidity and reference visibility may come from local instruments while precipitation context comes from the nearest available METAR.
Reference visibility remains validation-only
Visibility reported by a weather station or METAR can be retained for comparison and validation. It is not used as the VisSense Cam predicted visibility and is not allowed to directly choose the customer-facing visibility result.
Deployment requirement:
a camera without localized weather data must have known geographic coordinates before
the nearest-METAR fallback can be selected. Historical images must also be paired with
weather observations representative of their actual image time rather than current weather.
Camera deployment
VisSense Cam can adapt to different fixed cameras after calibration.
Cameras are not assumed to be optically interchangeable. Each deployment requires a camera-specific clear-scene reference and an independent commissioning check before equivalent performance should be assumed.
1 · Install a fixed camera
Keep the viewing direction, scene geometry, and zoom stable after commissioning.
2 · Collect clear-reference images
Use representative clear scenes to characterize the normal optical structure of that camera and location.
3 · Build camera calibration
Establish the camera-specific optical reference used to evaluate degradation in the current scene.
4 · Validate before operational use
Compare the new deployment against independent visibility information before claiming performance equivalent to an already evaluated camera.
Same-camera temporal history
Recent optical trend and change-point information must compare images from the same camera. Frames from another camera are not mixed into the temporal comparison.
Why camera-specific calibration matters
Optics, image resolution, orientation, distant landmarks, contrast, brightness, and clear-scene structure differ across deployment sites. Calibration reduces the risk of treating normal camera-to-camera differences as atmospheric visibility changes.
Operational output
Keep the operator display concise.
The annotated image communicates the current visibility interpretation while detailed routing, weather-source, and validation information can remain in diagnostics.
Example annotated output
StatusVERY LOW VISIBILITY <500 m
Estimated visibility<500 m
ConditionLIKELY FOG
Optical trendSTABLE
RH / Precip90.0% / NONE
UpdatedUTC timestamp
What each field means
StatusCurrent VisSense Cam visibility category.
ConditionInterpretation of the likely cause. “Likely fog” is intentionally not presented as a definitive fog observation.
Optical trendRecent same-camera optical change: improving, stable, or worsening. It is not a weather forecast.
RH / PrecipAtmospheric context representative of the camera site, using localized data first and fallback observations when needed.
UpdatedImage-analysis time in UTC, normally tied to the camera image timestamp.
Weather sourceDiagnostics can identify whether each weather variable came from a localized source or METAR fallback.
Operational examples
Example visibility interpretations.
These examples illustrate the operator-facing annotated-image format.
Very low visibility · likely fog
When reduced visibility is supported, precipitation is absent, humidity supports fog formation, and the camera shows compatible optical degradation, the condition can be reported as likely fog.
Validation
Camera transfer remains a commissioning and validation task.
VisSense Cam is currently a research prototype. The framework has demonstrated useful low-visibility detection within the available development and validation data, but a new camera or site should not be assumed to have equivalent performance solely because a calibration was created.
Camera-specific evaluation
Each new deployment should be checked using independent visibility observations and representative clear, fog, and precipitation cases.
Weather representativeness
Localized weather observations are preferred. When METAR fallback is required, the selected station and its distance from the camera should be retained in diagnostics.
Complement existing observations
The framework is intended to support situational awareness and research rather than replace certified operational visibility or aviation-weather instrumentation.
Interpretation:
low visibility is the detected state; fog or precipitation is the interpreted cause.
Reference-station visibility remains independent validation context.
About the research
Camera-based visibility research at the University of North Dakota.
VisSense Cam builds on camera-derived optical analysis, meteorological observations, machine-learning visibility modeling, and deployment-oriented weather research in the Department of Atmospheric Sciences at the University of North Dakota.
The current prototype is intended to support research on localized visibility awareness for aviation, UAS/AAM, roadway, and remote-site applications.