Slashdot Mirror
Engineers Say They've Created Way To Detect Weapons Using Wi-Fi (gizmodo.com)
An anonymous reader quotes a report from Gizmodo: The researchers, which include engineers from Rutgers University-New Brunswick, Indiana University-Purdue University Indianapolis (IUPUI), and Binghamton University, published a study this month detailing a method in which common wifi can be used to easily and efficiently identify weapons, bombs, and explosive chemicals in public spaces that don't typically have affordable screening options. The researchers' system uses channel state information (CSI) from run-of-the-mill wifi. It can first identify whether there are dangerous objects in baggage without having to physically rifle through it. It then determines what the material is and what the risk level is. The researchers tested the detection system using 15 different objects across three categories -- metal, liquid, and non-dangerous -- as well as with six bags and boxes across three categories -- backpack or handbag, cardboard box, and a thick plastic bag.
The findings were pretty impressive. According to the researchers, their system is 99 percent accurate when it comes to identifying dangerous and non-dangerous objects. It is 97 percent accurate when determining whether the dangerous object is metal or liquid, the study says. When it comes to detecting suspicious objects in various bags, the system was over 95 percent accurate. The researchers state in the paper that their detection system only needs a wifi device with two to three antennas, and can run on existing networks.
The findings were pretty impressive. According to the researchers, their system is 99 percent accurate when it comes to identifying dangerous and non-dangerous objects. It is 97 percent accurate when determining whether the dangerous object is metal or liquid, the study says. When it comes to detecting suspicious objects in various bags, the system was over 95 percent accurate. The researchers state in the paper that their detection system only needs a wifi device with two to three antennas, and can run on existing networks.
Unfortunately, from reading the paper, it seems like they've failed to measure the real effectiveness of their system -- they only measured the false negative rate and used just that for their "effectiveness". They never even bothered to check for or measure false positives. With such useless measures its easy to get a 100% effective system -- simply categorize everything as dangerous. Then your false negattive rate is 0% and your "effectiveness" is 100%.
Even 99% accurate is absolutely horrid for real-life situations. Let's take a case where one in a thousand items encountered are dangerous. I'm going to scale the numbers to make the results come out as integers. At 1 in 1000, if we start with 100,000 items we'll have 100 dangerous items and 99,900 non-dangerous. Of the dangerous items, the system will flag 99 as dangerous and 1 as non-dangerous. Of the non-dangerous items, the system will flag 98,901 as non-dangerous and 999 as dangerous. So out of 1,098 items flaged as dangerous, 90.98% of them will be non-dangerous. So a system that claims a 99% accuracy rate will have a 91% error rate when it comes to sounding the "Danger!" alarm. Only 9% of the time will that alarm actually indicate danger, the other 91% of the time it's a false alarm.
The above is why 5-nines (99.999% accuracy) is the baseline for workable systems.
I would think x-ray steganography would be a lot more useful. That is, unless you're arranging your contraband items in the shape of shorthand.
No need to panic, this is a method that uses a purposefully arraigned setup of wifi antennas to detect objects roughly 2 feet away, not some miracle technology that makes your home wifi able to scan your house. It's basically like cheaper version of an x ray machine at the airport monitored by an algorithm.