基于时序特征预测的实时视频拼接算法

Real-time Video Stitching Algorithm Based on Time-Series Features Prediction

在动态场景下进行视差处理是实时视频拼接中的一个难题,当运动对象经过视频拼接接缝时容易发生伪影和错位的现象,而复杂的配准和融合算法又难以满足实时性要求。为了解决这个问题,提出了一种基于时序特征预测的实时视频拼接算法。针对采集到的具有20%-40%重叠区域的视频帧,通过结合HOG和LBP等快速特征匹配算法,保证了视频拼接的速率。利用视频帧的时序上下文特性,识别出运动物体并估计物体运动矢量,判断物体能否到达接缝,选择性更新接缝,避免运动物体穿越接缝时产生畸变,达到了去伪影的效果。实验结果表明,所述算法视频拼接速率达到44FPS,且在去伪影效果上有很大的提升。

Parallax processing in dynamic scenes is a difficult problem in real-time video stitching.Ghosting and misalignment are prone to occur when moving objects pass through video stitching seams,and complex registration and fusion algorithms are difficult to meet real-time requirements.To solve this problem,a real-time video stitching algorithm based on time series feature prediction is proposed.For the collected video frames with 20%-40%overlap area,by combining fast feature matching algorithms such as HOG(Histogram of Oriented Gradients)and LBP(Local Binary Pattern),the rate of video stitching is ensured.Using the temporal context characteristics of the video frame,the moving object is identified and the object motion vector is estimated to determine whether the object can reach the seam,and the seam is selectively updated to avoid distortion when the moving object passes through the seam,and the effect of removing artifacts is achieved.The experimental results show that the video stitching rate of the algorithm reaches 44 FPS and the artifact removal effect is greatly improved.