Software To Help Eyes That See Too Much

United States military analysts are having some tough time since last 2-3 years. They are literally buried in analyzing the videos sent by their surveillance Drones. Drone is an Unmanned Aircraft System that generally contains an unmanned aircraft (UA), the control system and the control link. The surveillance drone has a camera that shoots the whole journey of the drone.

In 2009 alone, in Afghanistan and Iraq, the US Air Force shot over 24 years duration video using the spy drones. Now the main trouble arising is there aren’t enough human eyes to watch and analyze so much The deluge of this video data from these drones is likely to get worse!



This year the single new Reaper Drone will record 10 video feeds at a time, and the situation is about to get worse with Air Force planning to increase it to 65! So the duration of the video sent by the drones is about to increase exponentially and the existing strength of analysts won’t be able to analyze the whole thing. According to John Rush, chief of the Intelligence, Surveillance and Reconnaissance Division of the U.S. National Geospatial-Intelligence Agency, it would require about 16000 analysts to study the video footage from Unmanned Aircraft vehicles (UAVs).

The best possible way and I guess the only way is to have a computer system that will watch it all!  But programming a system that will automatically search all the videos and detect any wrong activity is not an easy task! So far, we don’t have software to perform this tedious task. But after 7^th IEEE international Conference on Advanced Video and Signal Based Surveillance, we have some hopes.

Mubarak Shah, director of the Computer Visions Lab at the University of Central Florida, in Orlando, presented three Computer Surveillance tasks that are very difficult.

Tracking Big Objects: Tracking Big objects, such as cars, is the first difficult task. When the aircraft like drone shoots from above, the cars look exceedingly small (usually not more than 30 pixels) and they are often in thousands! At the same time, the plane that is taking the footage is at a very high speed when compared to the vehicle being captured. So the tracking algorithm has very few frames for each vehicle.

Also, if the new technology being developed called ARGUS-IS succeeds, then a single drone mounted by the surveillance camera would be able to capture 1.5 to 2 million vehicles within a range of 40 Km during one mission. The solution given by Shah basically keeps the track of all possible paths the vehicle may have taken and then neglecting the poor choices. Things like Modern transportation theory (If one car is behind another, they are accelerating at similar rate) are used in the development of the algorithm.

Finding a Person: Obviously, if keeping track of vehicles is difficult, tracking a man is even more difficult. In an aerial video, the algorithms mistakes quite a lot of things such as trees, electric poles etc for people. Instead, Shah proposes of using some simple geometry tricks, based on relationship between height of the object and its shadow. But this solution won’t work on cloudy days.

Mapping the movement patterns: With many objects moving at once, mapping their movement patterns is very difficult. The computer surveillance system records a very noisy optical flow. Using mathematical noise-reducing tools, called Gaussian filters, we can find order in the noise and get the software draw a picture of the city in motion. We get to know the road networks without even knowing about the city.

The concept and the software are in their development stage. Lot of research needs to be done to get these things work accurately. For the military analysts, using these technologies as of now makes their job even harder. Using this new technology and accepting its results without going back to check the videos manually has a long way to go!