A Compound Eigenspace for Recognizing Directed Human Activities
This paper proposes a robust appearance-based method for recognizing directed human activities with scale variation based on a compound eigenspace. The method addresses two main issues associated with activity recognition when a human is moving away from or closer to the cameras. The first issue is the variation in human silhouette sizes as a result of object-camera distance changes. The second is the insufficient information of shape and speed of the limbs due to self occlusions. An eigenvector-based linear algorithm is employed for dimensionality reduction and activity recognition here. In addition to the conventional data available in each video frame, our method extracts two more pieces of information that are used to control the recognition process. In particular, the use of a compound eigenspace, controlled by the silhouette’s relative speed and linear displacement vector, has clearly improved the recognition. The method has been trained and tested using the four scenarios of the KTH dataset, which contains hundreds of videos partitioned into six human activities.
KeywordsHuman Activity Recognition Eigenspace Motion Intensity Image
Unable to display preview. Download preview PDF.
- 2.Diaf, A., Benlamri, R., Boufama, B.: An effective view-based motion representation for human motion recognition. In: International Symposium on Modeling and Implementing Complex Systems, pp. 57–64 (May 2010)Google Scholar
- 6.Jolliffe, I.T.: Principal Component Analysis. Springer (1986)Google Scholar
- 12.Schuldt, C., Laptev, I., Caputo, B.: Recognizing human actions: A local svm approach. In: The 17th ICPR, vol. 3, pp. 32–36 (2004)Google Scholar
- 13.Zivkovic, Z.: Improved adaptive gaussian mixture model for background subtraction. In: Proceedings of the 17th International Conference on Pattern Recognition, ICPR 2004, vol. 2, pp. 28–31 (August 2004)Google Scholar