Gyroscope Aided Blur Kernel Estimation for IR Image Deblurring

Stabilized Electro-Optical Systems are used in multiple domains, like target detection, tracking, and area surveillance. Cooled thermal imagers are used for long range surveillance. The requirement of a fast steering speed of the gimbal during surveillance induces blur in the captured IR video frames, which is required to be correct in real-time. Existing blind deblurring methods take a long time to execute. Deep learning methods are fast but cannot be generalised for all practical scenarios. In this work, we propose a noble blur kernel estimation method, using inertial sensor, and integrate it with existing non-blind deblurring methods and a sensor based deep learning method. Our method shows good results and is a suitable candidate for deblurring the image captured by gimbal-based imaging systems.