📷 Computer Vision for Bee Counting

About this simulation

Camera-based hive monitors mount a small lens above the landing board and run a lightweight object-detection model on every frame. Each bee that crosses the board is picked out as a bounding box, tracked frame-to-frame, and classified as moving inward (returning forager) or outward (departing forager). Counting these crossings over time builds an activity index — a proxy for foraging intensity that beekeepers use without opening the hive.

How it works

• The green translucent panel on the board is the camera's detection zone — the frame region the model actually scores.
• Bees crossing that zone get a live bounding box: green means the model's confidence cleared the threshold (a true detection), red dashed means it did not (a missed / false-negative frame).
• The monitor panel shows a rolling activity chart, exactly like the per-minute in/out counts an edge device would log.
• Low-light / IR mode simulates dusk or dawn conditions: motion blur and poor contrast lower effective confidence, so more bees slip through undetected — a real constraint that pushes hive-cam designs toward IR illumination or higher shutter speeds.

Controls

• Bee traffic rate — how many bees per second attempt to cross the board.
• Detection confidence — the model's confidence threshold; higher catches more bees but is more sensitive to poor conditions.
• Low-light / IR mode — toggles dusk lighting and a harder detection penalty.
• Show bounding boxes — toggles the live detection overlay.
• ↺ Reset counters — zeroes all counts. Drag to rotate, scroll to zoom.

Did you know?

Because raw video is far too heavy to stream continuously from an apiary with no reliable internet, most production bee-counters run detection directly on cheap edge hardware (a Raspberry Pi or similar) and upload only the resulting counts — a few kilobytes per hour instead of gigabytes of footage.