李盛楠 馬燕 黃慧 李順寶

摘 要： 提出了一種新的關(guān)節(jié)點權(quán)值自適應(yīng)的姿勢相似度計算方法，選用Kinect體感設(shè)備采集姿勢信息，獲取人體骨架關(guān)節(jié)點數(shù)據(jù).為適應(yīng)不同人體體型，根據(jù)骨架長度對關(guān)節(jié)點數(shù)據(jù)進(jìn)行修正.另外，針對不同的人體姿勢，提出自適應(yīng)的關(guān)節(jié)點權(quán)值定義方法.實驗結(jié)果表明：所提出的姿勢相似度計算方法準(zhǔn)確度高并且結(jié)果穩(wěn)定.

關(guān)鍵詞： 關(guān)節(jié)點權(quán)值; 源數(shù)據(jù)修正; Kinect; 權(quán)值自適應(yīng); 姿勢相似度

中圖分類號： TP 391.4? 文獻(xiàn)標(biāo)志碼： A? 文章編號： 10005137（2019）04035606

Abstract： A novel posture similarity calculation method using selfadaptive joint weight was proposed in this paper.Kinect was selected to collect posture information，using which the human skeleton joint data was acquired.In order to accommodate various body shapes，the data of joints was modified according to the length of skeletons.In addition，the definition of selfadaptive joint weight was proposed in terms of various human postures.The experimental results showed that the proposed posture similarity calculation method achieved high accuracy and stable results.

Key words： joint weight; source data modification; Kinect; weight selfadaptation; posture similarity

0 引 言

3 結(jié) 論

本文提出了一種新的自適應(yīng)關(guān)節(jié)點權(quán)值的姿勢相似度計算方法，該方法以模板姿勢關(guān)節(jié)點為基礎(chǔ)，對待測試姿勢的關(guān)節(jié)點進(jìn)行調(diào)整，有效解決了不同體型、位置之間的姿勢預(yù)處理問題.以模板姿勢的骨架長度為參考，給每個關(guān)節(jié)點增加一個權(quán)值，計算姿勢相似度.實驗結(jié)果表明：所提出的姿勢相似度計算方法效果較好.

參考文獻(xiàn)：

[1] SAHOO S P，ARI S.On an algorithm for human action recognition [J].Expert Systems with Applications，2019，115：524-534.

[2] MAITY S，BHATTACHARJEE D，CHAKRABARTI A.A novel approach for human action recognition from silhouette images [J].IETE Journal of Research，2017，63（2）：160-171.

[3] BEAUDRY C，PTERI R，MASCARILLA L.An efficient and sparse approach for large scale human action recognition in videos [J].Machine Vision and Applications，2016，27（4）：529-543.

[4] MOUSSA M M，HAMAYED E，F(xiàn)AYEK M B，et al.An enhanced method for human action recognition [J].Journal of Advanced Research，2015，6（2）：163-169.

[5] ASTERIADIS S，DARAS P.Landmarkbased multimodal human action recognition [J].Mutimedia Toolsans Applications，2017，76（3）：4505-4521.

[6] XU Z Y，ZHANG Q N，CHENG S Y.Multilevel active registration for kinect human body scans：from low quality to high quality [J].Multimedia Systems，2018，24（3）：257-270.

[7] SEDDIK B，GAZZAH S，AMARA N E B.Humanaction recognition using a multilayered fusionscheme of Kinect modalities [J].IET Computer Vision，2017，11（7）：530-540.

[8] LIU X P，LI Y B，LI X M，et al.Multifeature consultation model for human action recognition in depth video sequence [J].The Journal of Engineering，2018，（16）：1498-1502.

[9] GAVRILOVA M L，WANG Y X，AHMED F，et al.Kinect sensor gesture and activity recognition：new applications for consumer cognitive systems [J].IEEE Consumer Electronics Magazine，2017，7（1）：88-94.

[10] LI Q N，WANG Y F，SHARF A，et al.Classification of gait anomalies from Kinect [J].Visual Computer，2018，34（2）：229-241.

[11] THANH T T，CHEN F，KOTANI K，et al.Extraction of discriminative patterns from skeleton sequences for human action recognition [C]//IEEE RIVF International Conference on Computing and Communication Technologies Research Innovation and Vision for the Future.Ho Chi Minh：IEEE，2014：1-6.

[12] GIRSHICK R，SHOTTON J，KOHLI P，et al.Efficient regression of generalactivity human poses from depth images [C]//International Conference on Computer Vision.Barcelona：IEEE，2011：415-422.

（責(zé)任編輯：包震宇）