Posts Tagged ‘ video search ’

Monday, March 8th, 2010

by Paul Brewer

If you ask security directors how they go back into the video archives to search for something abnormal or for a specific event, the answer is almost always, “I don’t.”  Unless they know which camera feed to watch, and have a fairly good idea of the when to start searching in the video, this manual task is just too daunting. 

Video analytics has brought some much needed automation to this problem.  Now if you know where the event of interest occurred and can create a tripwire or other event rule, then all you have to do is to scroll through the alert logs and jump to that point in the video feed.

That’s great as far as it goes, but it’s still pretty limited.

What do users really want?  Based on direct user feedback, they want to be able to search for specific people, vehicles or events across the enterprise video system.  They want to use what they learn from one search to refine the next.  They want to search by example—designating a specific vehicle or person of interest to flag in a database search that might be narrowed by a specific geographic region or time period.  They want to create searches and visualize search results on an intuitive geo-interface.  And, they don’t want to just be limited to the video archives.  Forensic search results need to then become the parameters for real-time rules to find exactly where that white cargo van of interest is right now. 

This is the future of video search and it is what we are demonstrating right now for our sponsors in the Department of Defense.  A recent field exercise allowed ObjectVideo to showcase the ability to visually “fingerprint” cars that were flagged by HUMINT (human intelligence) and pick them out of the video feeds to present ultra high resolution snapshots to the “Battle Captain.” 

This was not a carefully controlled lab experiment with rigidly scripted scenarios with a small set of total vehicles.  This was a real needle-in-the-haystack exercise with a very small number of “opposition” vehicles operating on crowded public streets.  These vehicles were controlled by an opposition commander with purposes known only to him.

Besides being a great chance to show off what we can currently do with this technology, it was a priceless opportunity to learn from users and hone some new interface and workflow ideas.  When it comes to video search, the best days are still ahead, and possibly not as far off as you might think.

Tuesday, August 25th, 2009
by Paul Brewer

At least that’s how it goes in academia.

ObjectVideo researchers are offered the opportunity to publish their outstanding works in the proceedings of top computer vision conferences such as the IEEE Computer Vision and Pattern Recognition Conference (CVPR) and the IEEE International Conference on Computer Vision (ICCV).

The team’s latest publications describe some of the latest work that our research team has performed under funding from the US Army Tank Automotive Research, Development and Engineering Center (TARDEC) and the Office of Naval Research (ONR).

In the first paper [Geoffrey Taylor, Atul Kanaujia, Krishnan Ramnath, Niels Haering: “A Portable Geo-Aware Visual Surveillance System for Vehicles", in conjunction with the ICCV, Kyoto, Japan 2009] we see the benefit of integrating video analytics with an intuitive map-based interface to the camera systems that are increasingly being deployed on military vehicles.

The second paper [from Asaad Hakeem, Mun Wai Lee, Omar Javed, Niels Haering: "Semantic Video Search using Natural Language Queries", ACM International Conference on Multimedia, Beijing, 2009] explores the work that we are doing to develop the next-generation of video search applications, something that has tremendous potential in the commercial world.