Why KarlCam?
I made KarlCam because I wanted to learn but also because I found myself cursing the fog and its fickleness and was seeking a better way to avoid it. Here's roughly how it went down:
User Story
As a San Francisco cyclist planning a bike ride,
I want to see real-time ground-level fog conditions across different neighborhoods,
So that I can choose a route with sun or avoid areas where fog is at ground level.
Acceptance Criteria:
- • Weather apps only provide neighborhood-level forecasts without ground-truth data
- • fog.today shows satellite view but cannot distinguish clouds from fog. Fog is a cloud at ground level.
- • Need hyperlocal visibility data across San Francisco's varied microclimates
- • Must show current conditions, not just forecasts
Solution: Real-time SF fog visibility map using public webcam feeds
How it works:
- • Aggregates public webcams across SF neighborhoods
- • Uses computer vision to detect ground-level fog density
- • Displays color-coded fog conditions on interactive map
- • Updates every 15 minutes
Key features:
- • Live visibility status (not forecasts)
- • Neighborhood-level detail for SF microclimates
- • Route planning with fog avoidance
- • Mobile-friendly for checking before/during rides
Delivers: Ground-truth fog data so cyclists can choose sunny routes and avoid foggy areas
But wait... why do we need computer vision?
If Reed can see the camera images, what is the value of the computer vision classification of fog density?
The classification lets Reed scan the map to quickly understand where and how severe the fog is. Scanning lots of cameras isn't very easy, but if that's all Reed needed then an existing webcam map like Windy would meet Reed's need more simply.