Most computer vision nodes (the nodes which have a cvMat input & output) can be quite a drain on system resources, so care should be taken not to unnecessarily burden the system. In these chapters we’ll explain briefly what you can do to optimize the processes.

cvMat resolution

By far the most important parameter is the resolution of the cvMat that’s passing from one node to the next. In most cases, you can easily use very small resolutions, for example when there are less than 160 pixels in any direction. Feel free, of course, to experiment yourself.

texture resize node

If you want to reduce the resolution, if possible, use Texture resize node, not cvMat resize node as it is more efficient.

Tracking framerate

tracking framerate

You may want also to reduce the Tracking framerate in the properties of most computer vision nodes. For example, only very rarely should you increase Tracking framerate above 30. In a lot of cases, for example in bird’s-eye-view feeds of people walking, you can lower the tracking framerate down to 10 and still get results that are perfectly fine for your application.

Please note that you can reduce Tracking framerate of the first Texture to cvMat node and it will automatically affect all subsequent cvMat based nodes. The reason is that these nodes run only when the incoming cvMat actually updates.

Preview checkboxes

unchecked output preview checkbox

Having too many active cvMat based nodes with enabled Preview checkbox is also a resource drain because the conversion from cvMat to OpenGL texture consumes some resources as well.

print

Login to leave your feedback!

Comments are closed.