There is an option to use the SDK’s built-in module structure that is standalone, but also lets you extend or modify it, or create and insert your own algorithms and recognition subsystem to be run on the found faces. 

The default module structure of the SDK is shown on the image below.

The input images of the system may come from an IP camera using standard streaming and image formats. The most known formats are supported (RTSP, MJPEG etc.).

As it can be seen on the image above there is option to use different noise filters on the input images. We provide some examples, but it is easy to implement your own versions as well and use them as you wish. The image filters are applied one after the other, their order can easily be changed.

Different repositories are used to store information about the people appearing in front of given cameras and about given people’s appearance in the whole system.

The processing module is used to do multi-threading recognition on the input images. Our VisualArea SDK has its own person recognition system based on SURF point detection, of which multiple instances can run the same time. A special load balancer module has the task to feed the threads with data, during what its job is to even the threads as much as possible. Your new threads can also be put under the load balancer’s surveillance, if needed.

The results are processed by the output processor which sorts the incoming data packets by their content. Information about person appearance is saved into the repositories and the database as well, but there are options to extend this post-processing phase. An email sender interface is also available if you need to inform the supervisors about the alarms occurred.