Extraction of Multiple Motion Trajectories in Human Motion
This paper presents a method for extracting multiple motion trajectories in human motions. We extract motion trajectories of body parts (hands and feet) using a new method based on optical flow information. This procedure is not sensitive to complicated backgrounds or color distribution of scenes. No body part model or skin color information is used in our method. We first detect Significant Motion Points (SMPs) and obtain motion trajectories by connecting related SMPs through frames using Modified Greedy Optimal Assignment (MGOA) tracker based on the distance, motion similarity, and optical flow information. We test our approach on actual ballet sequences from videos. The resulting trajectories can be used as potential features for activity recognition.
- 7.D. Meyer, J. Posl, H. Niemann. “Gait Classification with HMMs for Trajectories of Body Parts Extracted by Mixture Densities,” The British Machine Vision Conference, Vol.21, no.10 pp. 961–973, Oct. 1999.Google Scholar
- 8.J.W. Deng, H.T. Tsui “An HMM-based approach for gesture segmentation and recognition,” In Proc. International Conference on Pattern Recognition, Vol.3, pp. 679–682, 2000Google Scholar
- 11.V. Kwatra, A.F. Bobick, A.Y. Johnson “Temporal integration of multiple silhouette-based body-part hypotheses,” In Proc. IEEE Computer Society Conference on Computer Vision and Pattern Recognition, vol. 2, pp. 758–764, 2001Google Scholar
- 12.E. Polat, M. Yeasin, R. Sharma “Detecting and tracking body parts of multiple people,” In Proc. International Conference on Image Processing, vol. 1, pp. 405–408, 2001Google Scholar