[ Potential applications of this research: Social networking in visual tracking, collaborative real-time communication, real-time visual behavior analysis ]
In 2010, I formulated a new research direction of social group discovery (SGD) for SSP  by a model-based approach using multiple features such as color, texture, motion, trajectories on detected humans. To the best of my knowledge, that was the first attempt to perform SGD  in a real world protest event with a sparse density of people (about 40-50 people in a scene in Fig. 1(a)) using a single feature (color appearance model using torso from detected humans in Fig. 1(b)) relying on a supervised method. A two-level of social network (Fig. 1(d)) was generated after SGD (Fig. 1(c)). A 94.34% success rate was obtained for SGD from that protest activity. It is believed that SGD can be further improved in aspects, such as utilizing a modified model-based approach with multi-features on detected humans, dealing with more complicated situation (more people in a scene), co-operating among multiple cameras, performing multi-level social network visualization, improving the algorithm to reduce the computational complexity, and developing an unsupervised method. I am still working on SGD in a larger dataset and would investigate the performance and reliability of the developed method.
 H.Y.T. Ngan, L. Wang and N.H.C. Yung, “Social Group Discovery via Two-level Social Network,” Proc. IEEE 3rd Int’l Conf. Power Electronics & Intelligent Transportation System (PEITS), Shenzhen, Chna, vol. IV, pp. 196-199, 20-21 Nov, 2010.
 H.Y.T. Ngan, H. Kawai, K. Kunieda, K. Yamada, “Social Behavior Analysis in Visual Human Monitoring System: A Survey and Perspective,” arXiv:1607.06219.