Encoding And Retrieval Of Motion Capture Data Based On Hashing Learning

Motion capture data is highly redundant in great magnitude scale and owns high-dimension feature space,which results in the long time consuming problem during data retrieval.In this thesis,hashing learning is applied for proposing a method of high efficient motion capture data encoding and content-based retrieval.Hashing algorithm owns desirable performance on high dimensional data retrieval.In this thesis,we place a sliding window containing a certain of frames of each sequence,it starts from the first frame and stops every few frames.We encode all the frames within the sliding window,which perfectly avoids the local sensitive of the motion actions.Hashing method is applied for data encoding,for the reason that the hashing code has shorter length and takes up less computer space.We proposed a hierarchical strategy to do retrieval,that is,first few representative frame is used for a rough retrieval of similar sequence,and then get the accumulated errors computed from the representative and sampling frames one by one,which is the key of the ranking.We argued from two aspects,the comparison of time and accuracy performance among the hashing and non-hashing method.It is demonstrated by the experiments that the proposed method has the better advantage for motion capture retrieval task,which inspires the motive of the management of motion actions.