Large Scale Video Data Mining

In today's real world video data are ubiquitous in almost all the aspect of the society and people's life. Given the typical large scale data collections, effective and efficient mining a large scale video data collection has become a daunting challenge. In this project, we address the following issues:

- video querying and retrieval  
- online real-time video target tracking  
- event detection in a massive video database.

Project Personnel: 





The work in this project is supported in part as fellowships awarded to the PI by the following foundations and the organizations:

- Chuo University, Japan

- Huang Kuan Cheng Foundation, Hongkong, China

- Japan Society of Promotion of Science, Japan

- Zhejiang University, China


The following demos are for the developed, tensor-based online and real-time tracking capability (called tensor in the demos) in comparison with a state-of-the-art tracking method (called PCA in the demos) under several different scenarios:

- Target with an abrupt motion

- Small target with a low contrast background

- Noisy quality video

The following demos are for the developed, real-time online video target detection capability under several different scenarios:

- Single target detection

- Multiple target detection



Go back to the Multimedia Computing Research Lab homepage