Small logo of ETH main building ETH Zurich : Computer Science : Pervasive Computing : Distributed Systems : Education : Student Projects : Abstract

Multiframe Blur Estimation using Inertial Measurement Sensors (B)

Status: Abgeschlossen


A common use case of smartphone cameras is to take photos of pin board advertisements or posters to remember a piece of information. The small form factor of today's smartphone cameras allows only a small aperture and hence longer exposure times. Camera movements such as handshake during the exposure result in motion blur in the recoded image.

The thesis deals with the removal of motion and handshake blur from pictures of flat surfaces with text such as posters, ads or price tags, using inertial measurement sensors and multiple frames from the video stream. The sensor values of the built-in accelerometer and gyroscope are evaluated to estimate the motion during a shot. This requires time synchronization of the camera and the inertial sensors. The information from multiple subsequent frames of the same scene with different blur can be exploited to reconstruct a sharp image of the scene.


  • Android NDK and OpenCV knowledge
  • You should have taken the courses Visual Computing and Distributed Systems


  • S. Cho, S. Lee - Fast Motion Deblurring, SIGGRAPH Asia 2009, link
  • N. Joshi, S.B. Kang, C. L. Zitnick, R. Szeliski - Image Deblurring using Inertial Measurement Sensors, SIGGRAPH 2010, link
  • S. Cho, H. Cho, Y. W. Tai, S. Lee - Registration Based Non-uniform Motion Deblurring, Pacific Graphics 2012
  • H. Cho, J. Wang, S. Lee - Text image deblurring using text-specific properties, ECCV 2012, link
  • Y. Li, S. B. Kang, N. Joshi, S. Seitz, D. Huttenlocher - Generating Sharp Panoramas from Motion-blurred Videos, CVPR 2010, link
Student/Bearbeitet von: Severin Münger
Contact/Ansprechpartner: Gábor Sörös

ETH ZurichDistributed Systems Group
Last updated September 10 2014 02:21:39 PM MET gs