加权最小二乘法与卡尔曼滤波实时稳像技术

无人机和车辆行驶等情况下拍摄的视频受外界影响会造成视频抖动。通过对比现有的电子稳像技术，提出了利用FAST获取特征点的位置信息，再通过光流法结合NCC匹配得到参考帧特征点在当前帧的位置信息，在此基础上，结合RANSAC算法剔除错误匹配的特征点对的改进算法。为了提高运动矢量估计的精度，应用加权最小二乘法得到相邻帧间的刚性变换矩阵，并经过卡尔曼滤波进行运动平滑得到扫描运动矢量并补偿，最终得到实时的稳定视频。实验表明，视频序列稳像后的帧间变换保真度有所提高，并且能够达到实时处理速度。

Improved least-squares and Kalman filtering real-time image stabilization

Video taken in the case of drones and vehicles traveling would be affected by the outside world causing video jitter. This paper proposed to use FAST to obtain the position information of feature points by comparing with the existing electronic image stabilization technology. Then it obtained the position information of the reference frame feature point in the current frame by the optical flow method combined with NCC matching. Based on this, it combined with RANSAC algorithm to eliminate the wrong matching feature points pairs. In order to improve the accuracy of motion vector estimation, this paper applied weighted least squares method to obtain the rigid transformation matrix between adjacent frames. The video was smoothed by Kalman filter to get the motion vector and compensated, and finally got a stable video in real time. The experimental table shows that it improves the fidelity of the inter-frame transform after video sequence stabilization, and achieves the real-time processing speed.