An Advanced Video Fusion System for Security Investigations
ITSP project (Project Reference : GHP/060/11) with 5M HKD funding.
06/2012 - 09/2014
(visualization for a steal sequence)
Video surveillance is deployed everywhere in HK to enhance public security. Security investigation is still much relied on checking 2D videos from separate camera views.
2D video segments do NOT provide global situational understanding (with both space and time registration).
- Each camera has a fixed viewpoint, a fixed orientation and a limited field of view.
- Hard to imagine the relative spatial position of people covered by different cameras.
Hard to find out the temporal order of activities/events within a large time window.
- Suspects may coordinate their activities by remotely communicating with each other with body languages or phones.
- Can not closely check an action or event from all viewing directions.
An internal 4D representation of past events.
- A simple pre-built 3D virtual model of an environment under surveillance.
- Recorded videos from a sufficient number of cameras.
- Recovered camera poses with respect to the 3D virtual model.
- On-demand object extraction and action reconstruction.
- Visualizing 3D individual or group actions over the 3D virtual model with a freely moving viewpoint.
- Prof. Yizhou Yu
- Dr. Bin Chan
- Yongtao Hu: Multi-target detection and tracking
- Wei Shao: target segmentation
- Guanbin Li: cross-camera tracking
- Frank Shen: User Interface
- Dr. Calvin Fong: 3D modeling & camera calibration, visualization
- Darren Li: Body Pose Tracking
- Dr. Forrest Qin: Body Pose Tracking
- slides (ZIP 74.1 MB, presented at Workshop on Video Analytics and Visualization for Security Applications on 2014-05-22)