High end video analytics over 2-4 Watt CPU

Proof of concept has been made this summer with High end video analytics Jaguar running over a 2-4 Watts CPU !

1. What makes Jaguar a high-end product ?

The Jaguar software is guaranteed to detect any moving target (visible on the image) of one or more pixels, and to raise less than 2 false alarms a day in sterile zones surveillance when cameras are installed according to our camera installation guide (cf. downloadable documents).

On all sites respecting this guide, where targets with a surface over 10 pixels are searched, the product was also proved this year to raise less than 0.25 false alarms per day.camera for installations with thermal cameras, and less than 0.35 false alarms per day.camera in case of color cameras.

False alarms are defined as any detection that does not concern a man sized target actually moving as specified in the sterile zones rules.

2. What actual performance was reached ?

The CPU target was an ARM9 CPU over a small sandbox card (~4x4 inches). The card consumption is ~3 Watts, without the processing running, and 4+ Watts when running. Several peripherals on the card can be removed (sound, accelerometer, etc), thus allowing to expect 4 Watts or even less on a specific board.  Images were acquired through a digital format (RTSP/RTP). The tests demonstrated that 320x240 pixels format images could be analyzed under a 160x120 pixels format at a framerate of 12 fps.

Such performance allows to detect 4 pixels targets (2x2 pixels in the original image, thus 1 pixel in the analysis format) at a framerate that fits well for thermal cameras (often limited to 8,33 fps).

The test also demonstrated that analyzing the true format in 320x240 is also possible at ~4 fps (good to detect relatively slow targets downto one pixel).

 

3. Future works

One of our next targets is ARM15 which will overcome largely these performances with a little more consumption. The other trend consists in cooperation agreements with camera suppliers who will integrate a camera open platform, ready for applications, with ARM9 or ARM15 processors.

 

Grants

These works were accomplished in the framework of a FUI R&D project of the SYSTEMATIC PARIS REGION Cluster, funded by the French MOD, through the DGA (Délégation Générale à l'Armement).