Study on the Particle-Filter-based Motion Filtering Algorithm for Digital Image Stabilization Systems

Liquan Dong,Desen Yin,Yuejin Zhao,Mei Hui,Xiaoxiao Zhou
DOI: https://doi.org/10.1117/12.834931
2009-01-01
Abstract:Towards to the existent problems in current motion filtering algorithms such as the low-pass filter, the Kalman filter and the extended Kalman filter. We proposed a novel motion filtering algorithm base on particle filter to separate random jitter from the global motion vectors. In this method, the variation of global motion parameters are regarded as the state-variable of system, the uniform motion model of camera is used, and the motion filtering is carried out to dynamic image sequences according to the features of digital image stabilization systems. Experimental results prove that this particle-filter-based motion filtering algorithm can achieve real-time filtering effect, and the filtering effect can be affected by the number of the particles only and almost irrespectively with other factors. This method can be used agilely and is very suitable for digital image stabilization systems application. We realized the programs based on the TI TMS320C6416DSP processing chip and got very perfect experiment results.
What problem does this paper attempt to address?