基于CUDA 和卡尔曼预测的实时电子稳像方法

针对传统电子稳像方法无法实现视频的实时处理的问题, 提出以SURF(Speed Up Robust Features)配准算法为基础, 基于CUDA(Compute Unified Device Architecture)编程实现算法的加速, 并利用卡尔曼预测器进行实时预测。算法利用CUDA 并行编程实现帧间特征点的提取和配准, 获得帧间运动矢量; 利用卡尔曼预测器获得稳定后的运动矢量, 实现对当前帧的运动矢量的补偿, 以达到实时稳像的目的。仿真实验结果表明, 该方法可有效去除视频帧间的抖动, 稳像效果良好, 实现了视频的实时处理。

Real-Time Electronic Image Stabilizing Approach Based on CUDA and Kalman Predictor

The traditional electronic image stabilization approaches cannot be realized in real time. For this reason, we proposed a novel method based on SURF ( Speed Up Robust Features) registration algorithm. Additionally, CUDA(Compute Unified Device Architecture) programming and Kalman predictor are used for algorithm acceleration and real-time prediction. The proposed method extracts, selects and registers feature points based on CUDA parallel programming, and then obtains the motion vector between video frames. The Kalman predictor is used to predict the motion vector of the current frame, and then to realize the motion compensation as well as the real-time image stabilization. The simulation results show that the proposed algorithm can effectively remove the shake between frames, obtain perfect image stabilization effect, and deal with the question of real time process.