I'm trying to build a robotic system that observes a moving object. In first phase of my project I have placed a simple black & white marker consisting of geometrical shapes on that target object. I'm able to estimate the pose of the object by locating and analyzing the geometry of the marker. However a problem arises when object moves a bit faster. This causes motion blur in the camera input. So far I have read Weiner, Lucy-Richardson and Blind Deconvolution techniques to remove motion blur. But they are very slow for a real-time application.

• Is there a faster solution to eliminate motion blur? I would prefer a library with python 2.7 interface.
• If not then how many fps are required to observe a human, who can run,walk, fall, jump without facing blurred images. What will be the cheapest option in such high fps cameras?
  

I've had the same problem. If possible increase your lighting so you can reduce the cameras exposure. If you cannot increase the amount of light can you open you F-Stop? You may also considerer a strobe light that fires when the camera is acquiring the image. Dave

I am not sure if you have options for camera hardware or not, but from the sounds of it you are using a rolling shutter camera, which would have the effect of introducing motion blur. Certainly increasing frame rate, and decreasing exposure time would have a positive effect on the image. However I would recommend that if at all possible switching to a global shutter camera which would do a lot more in reducing motion blur even at lower frame rates, and it would require no more computation on the computer.

Thanks @MackHaber and @doskey56
Reducing exposure worked perfectly for me. However OpenCV's native method to set the properties of VideoCapture object did not work for my webcam. So I had to use v4l2_ctl utility in linux to set exposure.