Research On Keytechnologies Of Electronic Image Stabilization Based On Multi-core DSP

Electronic image stabilization (EIS) technology is an important research area ofvideo processing and computer vision. It can improve the stability and definition ofvideo from cameras on moving carriers, and has been widely used in military affairsand civil cameras.This thesis presented a set of fast and parallel EIS algorithms. Firstly, in order tosolve the problem that the traditional motion estimation algorithm is too complex tobe applied to a real-time system, this thesis proposed a fast local motion vector (LMV)estimation algorithm based on the FAST feature. Secondly, a parallel global motionvector (GMV) algorithm based on image subarea is designed, which computes everyLMV simultaneously, and then estimates the GMV by all LMVs. Next, a motion filteralgorithm based on Kalman filter is presented to adapt to complex shootingenvironment and track camera motion, which separates camera motion from the jitter.Finally, the validaty of the proposed algorithm is tested on PC. Experoment resultsshow that the algorithm of this thesis can effectively restrain jitter componentof video.Based on the proposed algorithm, an EIS systems based on BF609dual-core DSPprocessor is designed and implemented. In order to achieve parallel execution ondual-core processor, this thesis designed a set of solutions about algorithm paralleltransplant, memory management, inter-core communication and cache consistency.Simultaneously, this thesis uses the pipelined vision processor (PVP) and Pixelcompositor in BF609to accelerate the algorithm and realize Gaussian filter and colorspace conversion.Experiment results show that the algorithm and system presented in this thesiscan effectively stabilize jittery video and achieve high real-time performance. Whendealing with PAL standard video, the total time cost, fixed relatively, is less than PALstandard frame period (40ms) and the system is reliable. The algorithm and system ofthis thesis are more suitable for the applications based on multi-core and fixed-pointprocessors. It is a viable EIS embedded solution of high practical value.