 |
Understanding Human Activity (P C Yuen et al.)
Understanding what people are doing from surveillance video cameras by machine is currently a hot research topic in computer vision and pattern recognition community because of many practical applications, ranging from national security video surveillance, shopping mall monitoring to domestic elderly smart home. To achieve this goal, we need to perform researches on (i) human detection and tracking within a camera view as well as across cameras, (ii) human action recognition and (iii) human-object/human interaction. The goal of this project is to understand what is happening in the surveillance scene.
Research Challenges:
-
Significant variations across different camera's views
-
Complex background with different lighting conditions
-
Appearance, pose and scale variations of humans
-
Difficulty in modeling human interactions and activities
Recent Findings:
- To address the tracking issue, we have developed an illumination insensitive face tracking algorithm [4] under single camera and a person re-identification algorithm for large scale camera networks [1]. Moreover, we have developed a salient detection algorithm [7] based on spatial and temporal information to identify some regions of interest in videos.
Illumination insensitive face tracking results [4]
Person re-identification (tracking) under disjoint camera views [1]
Spatio-temporal saliency detection results [7]
- For action recognition, we have developed a new representation, called eigenaction [6], for video representation. Based on the eigenaction and other features, we have developed a new co-training algorithm [5] to further boost the human action recognition performance. Since features are important for action recognition, we have further developed action feature learning algorithm using spatio-temporal neighborhood topology [2].
Generating Salient Action Unit by information saliency curve [6]
Salient Action Units in Weizmann database [6]
Top five EigenActions in Weizmann database [6]
Block diagram of the co-training algorithm [5]
Manifold learning based action recognition framework [3]
- Very often, multiple features are available and a typical way is to combine them directly with the assumption that they are independent. However, independent assumption is not true in many practical applications. As such, we have developed a dependency modeling framework for feature and classifier fusion [3]. It is shown that the performance can be improved in many visual recognition tasks.
Block diagram of score level dependency modeling [3]
Block diagram of feature level dependency modeling [3]
Publications:
- J H Ma, P C Yuen and J W Lai, "Domain transfer support vector ranking for person re-identification under target cameras without label information", Submitted to International Conference on Computer Vision (ICCV), 2013.
- J H Ma, P C Yuen, W W Zou and J H Lai, "Supervised spatio-temporal neighborhood topology learning for action recognition", IEEE Transactions on Circuit Systems and Video Technology, Vol. 23, No. 8, pp. 1447-1460, August 2013.
- J H Ma, P C Yuen and J H Lai, "Linear dependency modeling for classifier fusion and feature combination", IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 35, No. 5, pp. 1135-1148, May 2013.
- W W Zou, P C Yuen, R Chellappa, "A low resolution face tracker robust to illumination variations", IEEE Transactions on Image Processing, Vol. 22, No. 5, pp. 1726-1739, May 2013.
- C Liu and P C Yuen, "A Boosted Co-Training Algorithm For Human Action Recognition", IEEE Transactions on Circuit Systems and Video Technology, Vol. 21, No. 9, pp. 1203 - 1213, 2011.
- C Liu and P C Yuen, “Human action recognition from boosting Eigenactions”, Image and Vision Computing, Vol. 28, No. 5, pp. 825 – 835, 2010.
- C Liu, P C Yuen and G Qiu, “Object motion detection using information theoretic spatio-temporal saliency”, Pattern Recognition, Vol. 42, No. 11, pp. 2897-2906, 2009.
Grant Support:
This project is supported by the Research Grant Council (RGC), Hong Kong SAR, the National Natural Science Foundation of China (NSFC) and the Faculty Research Grant (FRG) of Hong Kong Baptist University.
For further information on this research topic, please contact Prof. P C Yuen.
 |