Event Cameras

A normal camera takes whole pictures at a fixed rate. An event camera has no frames at all. Each pixel works on its own and reports the instant its brightness changes. The result is microsecond timing, no motion blur, and almost no data when nothing moves.

Rotation speed Event threshold Normal camera frame rate

How a pixel fires

Each pixel remembers the log brightness at its last event. When the current log brightness $L$ moves more than a threshold $C$ away from that memory, it fires an event with its position, a timestamp, and a polarity: brighter (ON) or darker (OFF).

$$\left| \log I(x, y, t) - \log I(x, y, t_{\text{last}}) \right| \ge C \;\;\Longrightarrow\;\; e = (x,\, y,\, t,\, \pm 1)$$

What to notice

Spin the bar fast. The normal camera smears it into a blurry fan, because it moved during the exposure. The event camera shows two thin lines of events, red where the scene got brighter and navy where it got darker, right at the bar's moving edges. Stop the rotation and the event camera goes almost silent, since nothing changed.

A higher threshold means fewer events and less noise, but faint edges are missed. Event cameras suit fast drones and robots in tricky lighting. The catch is that most vision algorithms expect frames, so event data needs its own methods.