SmartVis Face has been designed to process video up-to Full HD video, i.e. 1920x1080. The minimum resolution requirement is driven by the minimum head-size in the video stream, which is 80 pixels. In general we recommend you analyse as few pixels as possible as this saves computation resource. This is why we recommend to set a Region-Of-Interest (ROI) on each video stream being processed.
Many cameras allow you to set the quality of compression used when transmitting the video stream. This is typically performed via the camera set-up interface. In general the higher the bits-per-second (bps), the better the recognition will be. For example, if you can see the block compression artefacts in the displayed video stream in Identifier, then this is a sign you need increase the bit-rate.
RTSP is a generic transmission format and many codecs can be used to compress the video. Typically, this can be configured using the camera configuration tool supplied by the camera manufacturer. We recommend H264 - this is typically the most stable and has been the one that is most tested with SmartVis Identifier.
The lens of the camera should be chosen to suit your application. You should set-up the optics of your system to concentrate on where the faces will appear in the image. We recommend that the faces in the image should not be smaller than 80 pixels in width. If the camera is far away, then change the focal-length of the lens to concentrate of the regions of the image the faces will appear. Identifier is best placed to observer pinch-points, i.e. entrances, exits, doors.
When the following points have been covered, you need to consider the operational set up considerations that are below:
- Enable the ROI
- Increase the bit rate, lower the fps if necessary
- Use H.264 codec