In the framework of a cooperation program in the energy domain, and with the help of a security integrator, we have deployed since March 7th, 2007 the EAGLE V1.3 product to verify its detection capacities and measure false alarm rates in a long period, as well as to establish alarm management protocols on "real data". We finally demonstrated a ratio of 1.6 false alarms per camera/day.

Settled in the east of France, the site is located on the top of a hill, in the countryside, close to a forest and private houses. There are foxes & fox families (once at the sunrise and once at the dawn), ravens (sometime fighting for some food item)... Storms, snow, fog have been observed on the site in the period. The site buildings are made of medium height installations (12 to 18 feet), with several high points (60 feet), surrounded by corridors of grass & vegetation (short lawnmowed). Surveillance areas are corridors around the site, half-yard long, watched by Samsung SCCB23xx/B20xx surveillance cameras, analog day/night with B&W / colour commutation. Camera gains (WB, AGC, ...) have been activated before our tests, we had to work without modifying the video architecture. The site is very dark at night, cameras are thus in a difficult situation.

A first observation period was led from March 7th to May 15th, concentrated on walking and crouching individuals. This campaign made it possible to measure and testify for an average daily false alarms rate of 4.5 per camera (by camera, from 4.2 to 4.7), due to animal life and weather conditions, and occuring mostly by day. Night surveillance was quite touchy in this period because of the very bad lighting conditions for these cameras (almost black image, blurred, under-resolved). An improvement in the lighting conditions in April improved slightly the situation for a formerly blinded camera, bringing it roughly to the same situation of the other cameras.

A new testing period was then driven with JAGUAR V2.0 between May 15th and summer, targeted on walking individual detection, and proved for an average false alarm rate of 1.6 per day & per camera. JAGUAR V2.0 is a new software implementing the latest researches from EVITECH, able to drive and manage several video flows in parallel (video matrix), and improved for video detection in very poor conditions. JAGUAR targets the civil markets, and sensitive sites protection.

The latest JAGUAR algorithms made it possible to demonstrate the excellence of the new approach followed by EVITECH R&D, on two critical aims :

• false alarm reduction through active filtering strategies,
• detection on very bad images through new stochastic learning algorithms.

On this last period, JAGUAR V2.0 proved its excellent detection capacity, including at night, and on depreciated images, in combination with a false alarm rate of 1.11 10-3 alarm per minute. The combined use of two independant detectors crossed on the same perimeter length with JAGUAR's reliability will thus provide a false alarm rate of 1,23 10-6 alarms per minute, and thus, with 525600 minutes per year, gives us 0,65 false alarm per year.camera.

Our customer on this pilot site feels, from these experimentation, very confident in the ability to reach a false alarm rate, which makes it possible to propose a new protection doctrine for its sensitive sites :

• double detection on video channel and physical detection (hyper-frequences, sismics),
• JAGUAR camera protection functions activation (detecting camera aggressions),
• alarm triggering on a combined alarm on both the video channel and physical channel in the same minute.

"Previous tests with a classical video sensor demonstrated to raise an incredible high false alarm rate", he said. "With a reliable solution like EVITECH's JAGUAR, we open the track for a high level automation of the security/intrusion function on our sensitive sites."