Evolution of video analytics started with motion detectors that simply subtracted pixel brightness of a frame from pixel brightness of the previous frame. Today the most sophisticated video analytics allow not just detecting moving objects, but also classify them with great precision.
Here’re some examples, showing the potential of modern algorithms.
Detection of items on a still background
Today we’re able to distinguish objects based on their shape. This works pretty similar to the way we, humans, do it.
Object detection on volatile background
Standard object detection works pretty well on a stable background. However constantly changing background made detection impossible. Today we’re able to solve this case. Please see this test: it’s hard to find a more harsh environment than ocean on a windy day.
Street video: detection in a poor light environment
Usual statement: avoid too dark, blurry and noisy images if you want to use detectors. Here we have pretty difficult conditions: some areas are so dark that it’s difficult to recognize anything with your eyes. However deep learning works well, detecting people on streets.
Did you notice a bicycle in the center of a group of people standing at the building’s corner? Video detector did!
Detection of numerous overlapping objects
Overlapping made detectors think of multiple objects as a whole. Here we detect each person separately.
Detection of still objects
Here furniture is well aligned with interior design. But the algorithm manages to distinguish it from the background.
Detection of cars from drone
Drones capture video from a different point of view. There’re usually numerous tiny objects to be captured in a single frame. The detector is able to capture most of them, calculate and classify them.
We at BitRefine group specialize in deep learning video detection and help companies solve the cases where standard detectors and approaches don’t work.