一种基于Star算法的多路视频稳像方法研究

Research on multi-channel video image stabilization method

摘要:

目前，基于特征点匹配的稳像算法中存在特征点提取速度慢、误匹配率高，以及多目相机的多路视频稳像太过耗时、实时性不好的问题，提出了一种基于Star特征点提取算法的多路视频快速稳像方法。首先采用Star算法提取视频当前帧特征点，然后采用光流法结合当前帧特征点信息跟踪预测下一帧对应的特征点，以加快特征点提取速度，同时减少所提取的特征点错误匹配率。通过对匹配的特征点求取相应的仿射变换矩阵获取帧间运动矢量，进而采用卡尔曼滤波器对帧间运动矢量进行滤波处理，以滤除其中的相机随机抖动矢量并保留相机的主观运动矢量。最后依据滤波后的运动矢量进行图像补偿得到稳定的图像序列。针对复眼导引头存在的多路视频抖动情况，基于OpenMP并行开发库，利用多核处理器的优势，实现了多路抖动视频的并行实时稳像。所设计的数字稳像算法在PC平台下对4路抖动视频（OTB100的BlurCar系列视频，分辨率为640×480，帧率为30 FPS）并行稳像的结果参数为：峰值信噪比 （Peak Signal-to-Noise Ratio，PPSNR）平均提升了4.62 dB，单帧耗时平均为14.51 ms。经仿真实验证实，算法的计算效率和特征点匹配正确率高，可实现多路输入视频的实时稳像。

Abstract:

Aiming at the problems of slow feature point extraction, high mismatch rate, and multi-channel video image stabilization of multi-camera cameras that are too time-consuming and poor real-time performance in current image stabilization algorithms based on feature point matching, this paper proposes an improved Star feature point fast extraction algorithm. First, the Star algorithm is used to extract the feature points of the current frame of the video, and then the optical flow method is combined with the feature point information of the current frame to track and predict the feature points corresponding to the next frame, which speeds up the extraction of feature points and reduces the number of invalid feature points extracted. The inter-frame motion vector is obtained by obtaining the corresponding affine transformation matrix for the matched feature points, and then the Kalman filter is used to filter the inter-frame motion vector to filter out the random camera shake vector and retain the subjective motion of the camera vector. Finally, perform image compensation according to the filtered motion vector to obtain a stable image sequence. In view of the multi-channel video jitter situation in the seeker, this article is based on the OpenMP parallel development library and uses the advantages of multi-core processors to achieve parallel real-time image stabilization of multi-channel jitter videos. The results of parallel image stabilization of 4 dithered videos (BlurCar series videos of OTB100, resolution 640×480, frame rate 30 FPS) under the PC platform of this algorithm are: PSNR increased by 4.62 dB on average, and single frame time consumption average is 14.51 ms. Experiments have confirmed that the algorithm in this paper has high computational efficiency and accuracy, and can realize real-time image stabilization of multiple input videos. Compared with other image stabilization algorithms, it has better practical value.