Analyzing gyro data based image registration

Abstract: An analysis of gyro sensor data with regards to rotational image registration is conducted in this thesis. This is relevant for understanding how well images captured with a moving camera can be registered using only gyro sensor data as motion input. This is commonly the case for Electronic Image Stabilization (EIS) in handheld devices. The theory explaining how to register images based on gyro sensor data is presented, a qualitative analysis of gyro sensor data from three generic Android smartphones is conducted, and rotational image registration simulations using simulated noise as well as real gyro sensor data from the smartphones are presented. An accuracy metric for rotational image registration is presented that measures image registration accuracy in pixels (relevant for frame to frame image registration) or pixels per second (relevant for video EIS). This thesis shows that noise in gyro sensor data affects image registration accuracy to an extent that is noticeable in 1080x1920 resolution video displayed on larger monitors such as a computer monitor or when zooming digitally, but not to any significant extent displayed on a monitor the size of a regular smartphone display without zooming. Different screen resolutions and frame rates will affect the image registration accuracy and would be interesting to investigate in further work. Ways to improve the gyro sensor data would also be interesting to investigate.