A Similarity Measure for Motion Stream segmentation and recognition.pdf

A Similarity Measure for Motion Stream Segmentation and Recognition ? Chuanjun Li B. Prabhakaran Department of Computer Science The University of Texas at Dallas, Richardson, TX 75083 {chuanjun, praba}@utdallas.edu ABSTRACT Recognition of motion streams such as data streams gener- ated by different sign languages or various captured human body motions requires a high performance similarity mea- sure. The motion streams have multiple attributes, and mo- tion patterns in the streams can have different lengths from those of isolated motion patterns and different attributes can have different temporal shifts and variations. To ad- dress these issues, this paper proposes a similarity measure based on singular value decomposition (SVD) of motion ma- trices. Eigenvector differences weighed by the corresponding eigenvalues are considered for the proposed similarity mea- sure. Experiments with general hand gestures and human motion streams show that the proposed similarity measure gives good performance for recognizing motion patterns in the motion streams in real time. Categories and Subject Descriptors: H.2.8 [Database Management]: Database Applications – Data Mining General Terms: Algorithm Keywords: Pattern recognition, gesture, data streams, seg- mentation, singular value decomposition. 1. INTRODUCTION Motion streams can be generated by continuously per- formed sign language words [14] or captured human body motions such as various dances. Captured human motions can be applied to the movie and computer game industries by reconstructing various motions from video sequences [10] or images [15] or from motions captured by motion capture systems [4]. Recognizing motion patterns in the streams with unsupervised methods requires no training process, and is very convenient when new motions are expected to be added to the known pattern pools. A similarity measure with good performance is thus necessary for segmenting and recognizing the motion streams. Such a similarity measure