SBIR Phase II: Multimodal Semantic Video Retrieval and Summarization

这个小企业创新研究第二阶段项目将开发上下文视频分割和自动标记技术和软件。在包含一个或多个主题的长视频流中，该软件自动发现每个视频中上下文连贯视频片段的开始和结束。此外，视频语义技术自动为每个片段分配文本标签，以便这些标签描述该片段中讨论的主题。所分配的标签使视频的所有部分都易于搜索。大型视频制作人目前依赖于手动将他们的内容分割成小段，并为这些片段分配文本标签，以使其可搜索。然后在每个片段之前插入一个简短的广告。这种手工分割和标记过程对内容生产者来说是一个重大的痛点，因为它是劳动密集型的，而且成本效益不高。与此同时，持续监控视频内容的政府机构依靠语音识别来识别特定的关键词。这种方法带来了两个痛点：(i)分析人员必须处理大量的错误检测，因为关键字出现的上下文可能是不相关的；（ii）如果关键字出现在重要的上下文中，分析人员仍然需要在视频中来回滚动以找到相关片段的开头。视频语义技术和产品有潜力有效地解决重大的市场需求。除了商业应用之外，拟议的技术将使媒体监测机构能够更有效地执行任务，节省宝贵的分析师时间和资源。此外，因为视频语义？技术是独立于语言的，媒体监测机构将能够监测更多的外语内容，而无需开发特定语言的技术。该公司将通过与开发媒体监控解决方案和元数据生成工具的软件公司合作，采用间接销售策略。该公司已经确定了它的第一个客户，并正在与他们合作，将上下文分割和标签技术集成到他们当前的媒体监控解决方案中。

This Small Business Innovation Research Phase II project will develop contextual video segmentation and automatic tagging technology and software. In long video streams that contain one or more topics, the software automatically discovers the beginnings and ends of Contextually-Coherent Video Segments in each video. Moreover, Video Semantics' technology automatically assigns textual tags to each segment such that these tags describe the topic discussed in that segment. The tags assigned make all parts of the video easily searchable. Large video producers currently depend on manually segmenting their content into small segments and assigning textual tags to these segments in order to make them searchable. A short advertisement is then inserted before each segment. This manual segmentation and tagging process represents a significant pain point for content producers because it is labor intensive and not cost effective. Meanwhile, government agencies, which continuously monitor video content depend on speech recognition to spot specified keywords. This approach inflicts two pain points: (i) analysts have to deal with large number of false detections because the context in which the keyword occurs might be irrelevant, and (ii) if the keyword occurs in an important context, analysts still need to scroll back and forth into the video to find the beginning of the relevant segment. Video Semantics' technology and products have the potential to efficiently address significant market needs. In addition to the commercial applications, the proposed technology will enable media monitoring agencies to perform their tasks more efficiently saving valuable analyst time and resources. Moreover, because Video Semantics? technology is language-independent, media monitoring agencies will be able to monitor more content in foreign languages without the need to develop language-specific technologies. The company will employ an indirect sales strategy via partnerships with software companies that develop media monitoring solutions and metadata generation tools. The company has identified its first customer and is working with them to integrate the contextual segmentation and tagging technology with their current media monitoring solutions.