dsutils | Command-line tools for doing data science

The first comment here is to learn how to visualize your DirectShow filter graph and understand what is effectively built. Whether you are developing in C++ or in C# through .NET interface library, you quite so often build by adding parts of the pipeline explicitly and then something is added for you. You don't know what you end up with exactly, so you need to…

… Understand Your DirectShow Filter Graph

That is, have your application build the pipeline and then inspect it with developer's tooling. This also helps sharing information when you ask a question like this.

In your case your graph building is not good: you first create a Null Renderer yourself, and then you request IVideoWindow and implicitly connect to Video Renderer supplied for you automatically. This means that you don't understand the pipeline and your question and problem is coming partially from this misunderstanding.

Next thing is that you are connecting live video source to synchronous renderer, which is - generally speaking - designed to display video with pre-buffered frames ahead of time. My guess here is that frame times slide away so that video renderer is blocking streaming and incrementally increases wait time causing more waiting and less capturing. I will skip details why it happens this way exactly, however important to know is that you are interested in inserting a Smart Tee Filter downstream to camera and upstream to renderer, connecting preview output to visual presentation. MSDN documentation gives some details on why this is expected to be done and the purpose.