视频剪辑查询结合时空金字塔匹配的视频检索方法

针对视频检索系统中目标持续移动从而影响检索精度的问题,提出一种基于视频剪辑查询融合时空金字塔匹配(spatio-temporal pyramid matching,STPM)方法.借助基于特征分析和分类的片段编辑检测器将新的视频分割成多个片段,以元数据信息将视频片段存入数据库,利用基于逐帧特征结合弱分类器的boosting算法检测视频片段边界,针对新的查询视频进行分析和线上视频匹配,并利用时空金字塔匹配计算相关反馈值.在中佛罗里达大学(university of central Florida,UCF)数据集和YouTube运动视频上的实验验证了方法的有效性,实验结果表明,方法的平均精度可高达97.6％,相比其他几种较为新颖的匹配方法,取得了更好的检索性能.

Video retrieval method based on fusion of spatio-temporal pyramid matching and video clips querying

As the problem of retrieval precision is impacted by continuous movement of objects in the video retrieval system, a spatio - temporal pyramid matching (STPM) method based on video clip querying is proposed. Firstly, the clips editing detector based on characteristics analysis and classification is used to divide new video into multiple clips, which are saved into database by metadata information. Then, boosting algorithm based on the characteristics of frames combined with the weak classifier is used to detect video clips border. Finally, analysis and online video matching are done for new query videos, and STPM is used to calculate relevance feedback value. The effectiveness of proposed method has been verified by experiments on UCF and YouTube active videos. Experimental results show that the average accuracy of proposed method can achieve at 97.6% . The proposed method has better retrieval performance than several other advanced matching approaches.