From Bits to Content : a Different Approach to Video Analytics

Prof. Pierre-Marc Jodoin
Department of Computer Science, University of Sherbrooke, Boulevard de l'Universite, Sherbrooke, (Quebec) J1K 2R1, Canada

Summary:
Video camera networks are ubiquitous. Over 30 million cameras produce close to 4 billion hours of video footage per week in the United States alone. This proliferation is taking place since, unlike other sensing modalities, visible-light cameras provide excellent space-time resolution, long capture range, wide field of view, and low latency. In that context, and with the ever decreasing cost of IP cameras, the need for simple but yet robust surveillance applications is becoming a glaring issue. Three such applications are (1) abnormality detection, (2) multicamera processing, and (3) efficient video-content retrieval. In this presentation, it will be shown that simple methods based on low-level bit-wise processing can lead to surprisingly accurate higher level results. A new video dataset devoted to motion detection will also be introduced. This dataset contains the largest number of indoor and outdoor hand-annotated videos in the world and propose a new way to evaluate motion detection methods. It is strongly believed that this dataset will have a significant impact on the video community.

Biography:
Pierre-Marc Jodoin is an associate professor at the Department of Computer Science of the University of Sherbrooke, Canada. He received the Bachelor degree from Polytechnique School of Montreal, and the M.Sc. and Ph.D. degree from the University of Montreal, Canada, all with honour. During his studies, he received 9 grants, all for the excellence of his work. His research interests include image processing, segmentation, 3D reconstruction, video analytics and surveillance. His recent work includes several anomaly detection projects, motion detection, segmentation methods applied to medical imaging, 3D reconstruction with unstructured light, and an image analytics software applied to the 3D movie industry (for which a patent is pending). His publication record can be found at the following address: http://www.dmi.usherb.ca/~jodoin/pub/publications.html. He also started in 2011 a company working in the field of medical imaging. The company is currently developing a software suite adapted to the needs of research in medical imaging and should start selling products in 2012.

 From Intelligent Robots to Non-intelligent Robots, A Pervasive Intelligence Approach

Prof. Ping Jiang
University of Bradford, England

Abstract:
Research on intelligent vehicle and mobile robot navigation has focused mostly on the development of a large and smart “brain” in order to gain autonomous capability copying homo sapiens. The approach is, however, facing a reliability problem and a computational bottleneck due to uncertainties in any dynamic environment. In this talk, the difficulties, especially in computer vision and navigation, will be discussed. It is argued that a machine should work in its own way, rather than just mimicking humans. This talk will report an intelligent environment with a mosaic of wireless camera eyes to support navigation and the control of mobile robots. The mosaic of camera eyes distributes the massive on-board intelligence required for autonomous systems to the environment. A robot with less intelligence can exhibit sophisticated mobility. The solution reported here uses a multiple Bloom-filter for the efficient storage of routing information and an active contour based scheme for path planning, trajectory generation, and motion control. A prototype intelligent environment consisting of 30 wireless visual sensors was developed for indoor navigation. The integrated experiments demonstrated the mobility of an environment-controlled wheelchair.

CV:
Dr. Ping Jiang is Reader in Robotics and Distributed systems at University of Bradford, England, where he leads research and teaching in intelligent robots and control. He received B. Eng., M. Eng. and Ph.D. degrees in Information and Control Engineering from Xi'an Jiaotong University, Xi'an, P.R. China, in 1985, 1988 and 1992, respectively. He was Lecturer in the Department of Electrical Engineering at Tongji University, Shanghai, in 1992 and promoted to Associate Professor, in 1994. Since 1997, he has been Professor in Department of Information and Control Engineering at Tongji University. His research work mainly focuses on intelligent control, distributed sensor and actuator systems, multi-agent and intelligent robotics. He has been involved in many academic research projects funded by
Chinese, British, EU and German foundations. In addition to academic research, he has hands-on experience on practical development and has been involved in more than twenty industrial knowledge transfer projects. From 1998 to 2000, he was an Alexander von Humboldt Research Fellow at Universitaet Erlangen-Nuernberg, Germany, where he worked on robot behaviour learning from demonstration and developed a universal learning approach for vision based imitation without camera calibration. From 2002 to 2003, he was
a senior research fellow in Computing at Glasgow Caledonian University for the IST Project DIECoM, where semantic web techniques were applied to distributed product configuration management. In his career, four robots were developed, which include a 5DOF educational robot (received the National Scientific and Technical Achievement Award in 1989 and a Silver
Medal in the 1990 Beijing International Industry Expo), a 4DOF direct-drive glass-cutting robot, a wheeled mobile robot and recently an autonomous wheelchair controlled by distributed visual sensors.

Using Complex Network Features and Wikipedia for Web Information Intelligent Processing

Prof. Ting Wang
College of Computer, National University of Defense Technology, P. R. China

Abstract:
With the rapid development of Web2.0 applications, the generating and diffusion of web information has changed greatly and many new features can be observed. For example, the Web information has tremendous scale and demonstrates many complex network features, the topics of content evolves very fast, and the web pages are jammed with noise or spam. These features bring great challenges to the Web information intelligent processing. This talk present an approach by using the complex network features and online knowledge base such as Wikipedia in the research of information processing on the web: (1) the topological features of complex networks has been used for fast clustering on web; (2) the social network between persons has been mined to improve the application of web person name disambiguation; (3) based on the semantic graph topic model using Wikipedia, a information diffusion tracking method can track dynamically evolving topic in noisy, short and multi-theme blog posts. The experimental results demonstrate that incorporating the network structural characteristics and knowledge can improve the web information processing.

CV:
Ting Wang received his Ph.D. from National University of Defense Technology (NUDT) in 1997. After graduated, he worked in the Department of Computer Science and Technology at NUDT as Lecturer and Associate Professor from 1997 to 2004, and then he was promoted to Professor. He has been involved as principal investigator in many research projects funded by the National Natural Science Foundation of China, the Program for New Century Excellent Talents in University, and the National High Technology Research and Development Program of China. Ting Wang is the author of over 60 technical papers in a wide range of conferences and journals including IJCAI, SIGIR, WWW, ISWC, ESWC and Journal of Web Semantics. His research work mainly focuses on Natural language processing, Information Retrieval, and Semantic Web.