Cobalt Video Event Detection

Detection under adverse conditions: heavy rain

Here we show the stability of the Cobalt Engine setups to variations in the ambient conditions. The same scene is analysed while it is sunny (left) and while it is raining (right). The setting of the Cobalt engine is identical in both. The scenes taken on different days from the same video camera looking through a window. The target is the house and the surrounding activity.

A street scene on a sunny day.

The same street scene on a very wet day.

The video control appears when moving the mouse over the video.
Using its video control you can view either video full screen. The ESC key returns to normal view.

The rain not only changes the illuminance of the scene: it also distorts the image as the rain falls on the window in front of the camera. The Cobalt Engine has to handle this distortion: the distortion should not cause events to be missed, nor should it cause false detections. This shows how Cobalt scene learning adapts to varying environmental conditions with no additional input from the user.

The Cobalt setup is identical in both scenes. Simple detection boxes have been placed at strategic points in the scene. No paramters needed to be set and the detection zones are completely independent of one another. This is the simplest possible setup.