Mechanized Attack Detection Algorithms

I was on a CNN sub site and they were talking about security and threat of attack. One of the types of attacks that they were profiling, was a mechanized terrorist attack by robot or remote control machines.

This got me to thinking. I am a technical architect for software, and its about time someone starting putting down some thoughts about algorithms for detection of mechanized attacks. Here are my thoughts on the subject.

1) The "Does This Thing Belong" algorithm. This algorithm is for machine monitoring of real time CCTV images. A camera aimed at a street would have a shapes learning module where each shape that went by the lense would be recorded. If a shape of an anomalous thing went by, an alarm would sound. A good analog of this is an electronic filter. You have high pass, low pass and bandwidth filters. In our algorithm, if you had a city street with no commercial traffic, then you would have a low pass object filter. Only small cars would be unalarmed. If a truck (possibly filled with explosives) went by, then the alarm would sound. The opposite of that would be in an industrial area, if a small vehicle was in an area where large trucks were the norm, an alarm would sound. The bandpass filter would detect things like robots, and remote controlled small vehicles sent against an urban target. Anything extremely large or extremely small would get noticed.

2) The purpose of movement filter would be next. Say you have a robot moving down the street, self navigating. One would expect some jitter in the movement. There would be small delays and corrections in self-navigation. However, if a CCTV camera was near a potential target, and it detected movement with no jitter or no evidence of a random walk navigation, it could mean that the direct route to the target was programmed in, and an alarm would sound.

3) The aerial threat probabilities calculator. If I were looking for aerial threats from airborne machines ranging from a toy helicopter with a hand grenade attached to a drone, I would create a huge software 360 degree virtual pillar around the sensors and camera, and then do a threat probability assessment of each sector of that virtual pillar. The highest probability sectors are from where the sun is the brightest (one of the tenets of air attack is to have the sun behind you to blind the observers). Then I would consider line of sight cover. Other factors that would either mitigate or enhance the risk would be no-go zones for an attack (a tall building is in the way). The threat probability from each sector would be calculated and monitored, and the system resources would be concentrated where the biggest threats lie.

4) Potential launch site evaluations. If one were to launch a mechanized attack, is there somewhere (either an entrance or egress) from where the attack would come. These areas would be heavily monitored.

These are just some of the algorithms to consider. The idea of a terrorist or other threat by robot or some other mechanized means is sort of remote now, but in ten years, I bet that this kind of software will be ubiquitous, and built into cameras. And you read about it here first.