Using VisualArea SDK in your own environment is not too difficult. Our software developement kit arrives as a collection of .jar and .dll files (.so for Linux users) that must be included in your software and then it is ready to be used.

IP video cameras should be used; you will need the whole URLs of the streams and use them extended with the port number, username and password.
E.g. rtsp://user:pw@192.168.0.1:554/stream.sdp

Training (learning people's faces) are done by extracting data from faces in a given folder. Read more details in the Learning Faces section.

VisualArea SDK supports multi-threading and this should always be considered during your development. Be careful and implement thread-safe solutions only, and create such code that does not throw exceptions because of concurrent modifications.

The minimum hardware highly depends on your exact usage, but image processing is rarely a light task even for the state-of-the-art processors. We suggest at least 8 GBs of system memory and a dedicated server.

An example calculation to help a bit: if you are about to use a database of 50 people learnt on two stream cameras and you would check personality twice per second, face comparisons must be done within 200 times a second! Depending on the exact hardware, the resolution, the number of SURF points and other factors the speed of one face-check is usually between 50-200 ms. Let’s calculate with the former, then 1000/50 = 20 calculations per core is reachable, in other words at least 10 processing cores are needed, but at least 2 other are used for stream reading and stream manipulations, so the final number is 12.

If thinking about using multiple high-resolution cameras with bigger frame rates, think about the needed enormous network traffic as well. We suggest a resolution of 1024x768 pixels and a frame rate of 2-5 fps. Remember, it’s not a television series, but security camera streams.

In summary first of all the number of processing cores is important (the more the better), other factors such as memory are not that crucial.