The detection system locks on objects in a running MPEG-2 stream without fully pixel level decoding. (clockwise frames 0, 50, 350, & 400 ). (Top) Real-time tracking of objects as they appear and disappear in the scene (objects 1 coverag effectiveness (blue) and miscoverage (purple). |
Coding Depth Sensitive Perceptual Video Interceptor |
With the reduction of cost of cameras scenarios are contemplated that would require hundreds and thousands of cameras to be deployed.
It seems a new age of ubiquitous video processing is in the horizon. A mile stretch of highway may have few hundred cameras. Interior spaces such as an airport observation system can have denser deployment as high as 5,000-10,000 cameras in public areas. It is humanly impossible to watch such massive regimen of cameras.
Interestingly, most of today’s video object detection techniques have been historically derived from image processing techniques and thus require access to fully decoded pixels and thus painfully slow. Even the fastest algorithms of the group are order of magnitude inadequate to meet the challenge of the onslaught of this video age.
In this work we are investigating a new generation of video object detection algorithm which is specially designed for complex coded video.
Based on some recent results in human visual perception, in this research we present experimental visual region tracking algorithm particularly designed for perceptual stream transcoding.
|
>> MMN: Javed I. Khan and Zhong Guo, Fast Perceptual Region Tracking with Coding-Depth Sensitive Access for Stream Transcoding, Elsevier Journal of Visual Communication and Image Representation, JVCIR (accepted, October 2007).
>> MM: Khan Javed I, Motion Vector Prediction in Interactive 3D Video Stream, Proceedings of the World Congress on Advanced IT Tools, IFIP ‘96 IT, Canberra, Sept 96, pp533-539. [PDF]
>> MMN: Javed I. Khan and Darsan Patel, Extreme Rate Transcoding for Dynamic Video Rate Adaptation, Proceedings of the 4th International Conference on Wireless and Optical Communication, WOC03, July, 2003, Banff, Canada. [KhPa03], (GS(1)).
>> MMN: Javed I. Khan and Zhong Guo, Flock-of-Bird Algorithm for Fast Motion Based Object Tracking and Transcoding in Video Streaming, Proceedings of the 13th IEEE International Packet Video Workshop 2003, Nantes, France, April 2003.[PV2003] [KhGu03] |
Selected Publications |
More Publications >> |
Rather than begin processing from pixel level or using any pixel level processing at all, it employs high level motion cue and block shape cue analysis to identify signatures of various relative movements between object of interest, scene background and the camera on the motion vector set, and from there it identifies objects. It then uses predictive filters to track the regions of interest.
The result is a fast yet highly effective perceptual region tracking algorithm that can operate in stream rate and track regions of perceptually significant object despite camera movements such as zoom, panning and translation.
We have also developed techniques where even re-quantization and transcoding can also be performed with coding depth sensitive scheme, without requiring the stream to be fully decoded for blazingly fast near transparent video intercept. |
Publications Distributions Demo Technical Reports All Projects
|
Page last updated February, 2008, Medianet Laboratories. |
|
Welcome | Projects | Publications | Technical Reports | Software | Resources | Sponsors | Personnel |