Research On 3D Human Motion Classification And Recognition

Nowadays motion capture technology has been widely used. With those motion capture equipments, Millions of 3D human motion data are obtained and have been used in a lot of areas, such as computer animation, games, entertainments, etc. There is an increasing demand for better 3D human motion classification and recognition considering that it's the fundament of reusing these human motion data.This paper focus on two technologies:3D human motion classification and 3D human motion segmentation.This paper proposes a human motion classification method based on sparse representation. Firstly, an overcomplete dictionary needs to be computed from training data. Then each frame in the test sequence is treated as linear combination of a few basic elements of the overcomplete dictionary, and its sparse representation is computed by L1-minimization. Finally each frame is classified by computing residual.This paper proposes a human motion segmentation method based on spectral clustering. Given a long series of human motion, a graph containing timporal information is constructed and then a spectral clustering method is applied to guarantee the accurate segmented result.The result demonstrates that our method outperforms other traditional methods. We exploit this method to process motion segmentation and the result turns out to be satisfactory. Meanwhile, our segmentation method based on temporal spectral clustering achieves wonderful results.