动态场景下基于VR技术的短视频实时分割

Real-Time Segmentation of Short Video in Dynamic Scene Based on VR Technology

针对现有短视频分割方法中存在应用场景受限与运动背景过分割等问题,提出动态场景下基于VR技术的短视频实时分割。利用VR技术三维重建动态场景下的短视频,丰富短视频的细节信息;通过时空联合的视频帧对象分割方法分割三维重建后的短视频,通过最大帧差分量方法提取连续两帧的帧差信息,利用线性扫描填充方法填充获取帧差模板;通过融合模糊C均值聚类算法与运动窗技术剔除帧差模板内的背景,获取对象模板;通过融合区域生长与边缘检测方法填补对象模板,获取完整目标对象;轮廓修正完整目标对象,获取精准目标对象,完成短视频实时分割。仿真结果表明:所提方法对短视频的分割精度较高,且分割效率较高。

Traditional short video segmentation methods have many defects,such as limited application scenes and over segmented moving background.Therefore,the short video real-time segmentation based on VR technology in dynamic scene is proposed in this work.VR technology was applied to reconstruct short video in dynamic scene,enriching the details of short video.Based on spatiotemporal joint video frame object segmentation method,the short video after 3 D reconstruction was segmented.According to the maximum frame difference component method,the frame difference information of two consecutive frames was extracted.The linear scan fill method was used to obtain the frame difference template.Combined with fuzzy c-means clustering algorithm and moving window technology,the background in the frame difference template was excluded to obtain the object template.According to the fusion region growing and edge detection method,the object template was filled and the complete object was obtained.The complete object was modified by profile,and the accurate object was obtained,thus completing the real-time segmentation of short video.The simulation results show that the method has high segmentation accuracy and efficiency.