中国现有听障人士超2700万，为了他们能够“倾听”冬奥会的一系列播报，央视推出了首位AI手语主播，人工智能网络与手语研究的结合已经小有所成。根据研究表明，中国手语视频数据集大多是由手语短语或短句视频组成的，很少有中国手语词汇级视频资源库，为此，我们设计并完成了一个中国手语视频数据库词标注众包平台。

本文通过借助众包平台里接包方的力量实现专业领域的视频分割标注并减小误差，在视频自动分割标注上采取了以视频帧为单位的分割方法，使用众包来提高了视频分割标注的准确性。实验结果表明，借助众包方法可快速建立和完善词汇级的中国国家通用手语视频数据库，以视频帧为单位选择的分割方法极大提高了众包结果的准确性，为进一步研究建立手语学习模型算法，实现手语合成提供了条件。

With more than 27 million people in China with hearing impairment, CCTV has launched the first AI sign language anchor to enable them to “Listen”to a series of broadcasts of the Winter Olympics. According to research, the Chinese sign language video dataset is mostly composed of sign language phrases or phrase videos, and there are very few Chinese sign language vocabulary-level video resource libraries, for which we have designed and completed a Chinese sign language video database word annotation crowdsourcing platform.

In this paper, by using the power of the subcontracting party in the crowdsourcing platform to achieve video segmentation labeling in the professional field and reduce the error, the video automatic segmentation labeling adopts the segmentation method based on video frames, and the accuracy of video segmentation labeling is improved by using crowdsourcing. Experimental results show that the vocabulary-level Chinese national general sign language video database can be quickly established and improved by means of crowdsourcing method, and the segmentation method selected by video frames greatly improves the accuracy of crowdsourcing results, which provides conditions for further research and establishment of sign language learning model algorithms and the realization of sign language synthesis.