Motion-invariant photography

Motivation Scenes with moving objects are difficult to image especially with high exposure due to motion blur. One option is to segment the image into different regions based on moving objects and deblur each of them individually. But this method may lead to artifacts in the edges between objects (if at all, segmentation is perfect). How to do away with the segmentation and space-variant deblurring?

Track objects If the blur is space-invariant, then the deblurring will output better images that that of space-variant. If we keep the camera static in a dynamic scene, the blur will always be space-variant (assuming different objects move with different velocities). The important observation to make here is that the background will appear clean in this case, since its velocity is zero. In other words, the camera is tracking the background (since both their velocities are equal, yes, zero). If we move the camera with a velocity \(v\) in a dynamic scene, then objects moving with velocity \(v\) will appear clean. Thus, if we could track all objects within a single exposure time, it would be imaged clean at some point in the exposure. But, what is its use?

Make it space-invariant Consider you are looking at a road with many vehicles. The road will look clean to your eyes, and the fast moving cars will look blurred. Now, shake your head fast, left and right continuously. Now, the road also will look blurred in addition to cars. But, the blur is same for all objects and the road. That is, if you want to deblur easily, blur more. Counter-intuitive?!

Sweeping the camera If we know the range of possible velocities of objects in the scene, say \([-vmax,vmax\]), then we can sweep the camera through all these velocities during a single exposure time. This would create a space-invarnt blurred image

//explain more//