Analysis Of Dance Movements Based On Multi-feature Fusion

Recently,music and dance video analysis has been favored by researchers.Through the effective analysis of the music and dance videos,we can correct the dancers' dance posture,assist the choreography and aid the dancers in training,which plays a certain role in promoting the development of the music and dance.However,due to the complex movements and repeated actions in videos,the difficulty of content based video clip retrieval is increased,which is difficult to satisfy users' requirements for video retrieval accuracy.Therefore,this paper presents a dance video clip retrieval method based on video summarization,the main contents are as follows:A key frame extraction method for music and dance video is proposed.The first step is to divide the frames of the dance video.After that,the optical characteristics of every danced image are calculated to get the motion characteristics of the dancers.Then the entropy of the optical flow is calculated to obtain the entropy sequence.The accompaniment music in the dance video is used for audio feature extraction to obtain the audio feature sequence.Then the above two sequences are fused to obtain a sequence of entropy values based on the change of the audio frequency.Finally,the sequence is used to extract the key frames,and in the process,the optimal threshold is selected through continuous iteration,so as to obtain a high quality set of key frames.A video clip retrieval method based on video summarization is proposed.First of all,the key frame sequence of the video is extracted,and the key frame is used to represent the content of each video clip.Afterwards,the ORB feature is used as the feature of the video clip.The time sequence factors,granularity factors,interference factors and other factors are taken into account,when similarity matching model is built.A video clip that is similar to a query clip is selected by comparing similarity.The experiments are conducted on two dance datasets in this paper.One is a public dance dataset,and the other is recorded by a Vicon motion capture system in the laboratory.The evaluation operation is performed on the video clip retrieval of the music and dance proposed in this paper.The experimental results show that the proposed method can retrieve the video clips of the users' requirements under the condition of guaranteeing the rate and accuracy.