100% Video Analytics Accuracy?By: Ethan Ace, Published on Feb 01, 2012
In January 2012, analytic provider SightLogix announced its SightSensor intrusion detection analytics had been validated by the Transportation Safety Administration, providing "100% video analytics accuracy", a bold claim. In this note, we look at this announcement, the TSA test, the integrator's experience and what users should look for when reviewing manufacturer test reports.
While the report [link no longer available] was prepared by the TSA, the evaluation of the SightLogix system was performed by the Safe Skies Alliance, which is a non-profit formed in 1997 to support testing of aviation security technology. The SSA receives funding from the FAA, and support from the TSA. This testing was performed at Buffalo International Airport in March 2011.
SightLogix has said that this report will be available to other airport security departments, but not to the public. While we understand the need for redaction, and keeping some of these measures secret, any number of other industries could benefit from this test, as well. As one commentor pointed out, though, it would take TSA involvement to distribute this to even industries outside of aviation, let alone the security industry at large.
SightLogix's perimeter detection system is well respected and higher cost than many other options in the industry. It has also been developed for the specific purpose of intrusion detection, unlike many all-in-one analytic packages (see our review of their new 2012 camera series). Given this, we would not be surprised if detection rates are high, with false alarm rates low. However, users should be careful when reading this test report and press release, as some claims may be misleading.
- Location and coverage information is redacted, so we're left with no knowledge of how wide the area covered is, nor are detection distances shared.
- Intrusion techniques (the scenarios) were redacted, so we've no idea if subjects were detected while walking or crawling, from which direction they approached, at what speed, etc. Results for each of these scenarios were also redacted, so performance by intrusion type is unknown.
- All nuisance and false alarm reports are redacted. It is extremely unlikely that any systems can be free of false and nuisance alarms, so this lack of detail removes information regarding potential system weaknesses.
In a LinkedIn discussion [link no longer available] regarding this release, one of our members who was involved with the integration shared the following clarification:
- Regarding false alarms: "We had a number of alarms in the beginning and by a process of adjusmtments we were able to tune the system for the customer to an acceptable level. That is the key when installing an analytic system is setting the expectations of the customer correctly. 99% of the alarms are caused be animals and dependant on the size and quantity may or may not be a concern for the airport."
- Regarding Nuisance alarms: "It does track objects that are not humans, but that was an added benefit to certain airport operations as birds, deer and other animals are not wanted guests on a runway. The biggest benefit to the system is that if there is an alarm it is automatically time stamped and logged, a monitor displays the target and the police officer can, by not touching anything, view to see if the alarm is something that requires attention."
- Regarding performance: "I can say that the system after being in service for over a year is still providing a useful tool that is reliable and paying for itself."
He also stated that the system has been expanded to a second airport, bus garages, and a light rail system, which would lead us to believe the system is performing satisfactorily.
We found the integrator's insights to be far more useful than the study itself. A key factor was that the integrator was able to share experiences in the optimization stage and how it is used in daily operations. The lesson to take away - If you want a good reference, stay away from 'studies' and find an engineer who was involved in the details.