基于人脸角度估计的多观察视频合并

Combination of multiple observation videos based on face angle estimation

对特定人物进行观察记录时,由于被观察对象处于运动状态,需要通过布置多个观察源来充分采集其行为信息,这导致大量冗余信息的存储,同时不利于后续对视频的检索和对被测对象行为的分析。为此,提出了一种基于人脸角度估计的多观察视频合并方法,通过对多个观察视频的拆分、人脸检测、人脸角度估计和重组,获得单一的包含被观察对象正面行为的观察视频,然后进行存储。通过实验验证了算法的可行性,同时讨论了多观察源采集视频不同步带来的影响。

While observing and recording specific people,multiple videos are needed for sufficient information gathering,which leads to storage of abundant useless data and results in difficulty for searching and analyzing of observation videos. A combination method of multiple observation videos based on face angle estimation is proposed to solve this problem. A single video which only contains facing scenes of moving people is obtained and stored by splitting multiple observation videos,detecting face and recombining based on the face angle estimation.Experiments verify the feasibility of this method and the influence of asynchronous videos is discussed.