发明名称 Representative motion flow extraction for effective video classification and retrieval
摘要 Technologies are generally described herein for extracting a representative motion flow from a video. Technologies are also generally described herein for retrieving a video utilizing the representative motion flow. The representative motion flow may be extracted utilizing a sliding window approach to generate interesting motion flows. The representative motion flow may be generated based on the interesting motion flows.
申请公布号 US9268794(B2) 申请公布日期 2016.02.23
申请号 US201514602378 申请日期 2015.01.22
申请人 PEKING UNIVERSITY 发明人 Cui Bin;Zhao Zhe
分类号 G06F17/30;H04N5/232;G06T7/20;G06K9/00;G06K9/46 主分类号 G06F17/30
代理机构 Turk IP Law, LLC 代理人 Turk IP Law, LLC
主权项 1. A computing device configured to extract a representative motion flow from a video, the computing device comprising: a memory; a processor coupled to the memory, the processor executing a video retrieval application, wherein the video retrieval application is configured to: receive a query video file;extract the representative motion flow from the query video file through a process by: dividing the query video file into sliding windows, wherein each of the sliding windows includes at least two frames;for each of the sliding windows, generating motion flows,clustering the motion flows that share one or more of a pattern of the query video file and a location of the query video file into a first subset of clusters,extracting first meaningful clusters from the first subset of clusters, andmatching the motion flows in the first meaningful clusters of a current sliding window of the sliding windows with the motion flows in the first meaningful clusters of previous sliding windows of the sliding windows to generate interesting motion flows; anddetermining at least the representative motion flow based on the interesting motion flows;divide the representative motion flow into segments through use of a bottom-up segmentation technique, wherein each of the segments corresponds to one or more elements of trajectories of objects within sequential frames in the query video file, and wherein the bottom-up segmentation technique groups pixels of homogenous segments of the representative motion flow according to a texture uniformity of the homogeneous segments;map a set of symbols to each of the segments to generate a query symbolic representation of the representative motion flow, wherein the query symbolic representation depicts the trajectories of the objects within the sequential frames in the query video file,wherein each symbol in the set of symbols represents a direction of the segments in the trajectories, andwherein a number of symbols in the set of symbols represents a length of the segments in the trajectories; andretrieve candidate video files with corresponding candidate motion flows that have candidate symbolic representations that match the query symbolic representation based on a level of matching between the candidate symbolic representations and the query symbolic representation, wherein the level of matching includes one of a complete matching, a substantial matching, and a partial matching.
地址 Beijing CN