一种改进的DTW算法在人体行为识别中的应用

目前针对各维特征之间的相关性对动作识别影响的问题，解决的方法基本是采用欧氏距离作为其相似 度的算法.结合Kinect体感设备提出了一种基于改进的DTW算法的人体行为识别方法，将马氏距离作为相似度 测度引入DTW算法中进行改进.首先，利用Kinect设备采集人体骨骼信息，计算骨骼夹角信息，之后使用改进 的DTW算法进行模板训练和人体行为识别.最后通过实验，对比采用欧氏距离、卡方检验和马氏距离作为相 似度测度时，人体行为识别的准确率.实验表明引入马氏距离的DTW算法在识别正确率方面有所提高.
AhumanbehaviorrecognitionmethodbasedonanimprovedDTWalgorithmisproposedbycombiningthe Kinectsensorinthepaper.Tosolvetheproblemthatifthecorrelationbetweeneveryfeaturewillaffectbehaviorrecogni? tion,Euclideandistanceisadoptedintheexistingwayasitssimilarityalgorithm.Mahalanobisdistanceisintroducedin? totheDTWalgorithmassimilaritymeasure.Firstly,weuseKinectsensortocollectthehumanskeletoninformation;then theimprovedDTWalgorithmisadoptedfortemplatetrainingandhumanbehaviorrecognition.Finally,compareaccura? cyrateofhumanbehaviorrecognitionwhenEuropeandistance,Mahalanobisdistanceandchi-squaretestseparately serveasthemeasureofsimilaritydetection.TheexperimentsshowthattheDTWalgorithmintroducedbytheMahalano? bisdistancehasbeenimprovedintherecognitionaccuracyrate.