This is as much as possible what the Camera cue in QLab[1] does (I'm the lead video developer), and there most definitely is perceived latency. It's better than it used to be, when webcams were DV over Firewire, but it's still enough that webcams are generally a poor choice for latency-sensitive situations like live performance.
The alternative that usually gets used in productions with a moderate budget is a hardware camera with HDMI or SDI output into a Blackmagic capture device. There's still some latency, but it's better than a webcam. It's pretty much down to the minimum of what you can do with a computer -- the capture device still has to buffer a frame, transmit it to the computer, which copies it into RAM, then blits it to video RAM, where it waits for the next vertical refresh.
All that said, the human finger doesn't travel all that fast, so for a HID application like this, you could probably get the latency way down by extrapolating the near-future position of the finger, much like iPadOS does with Pencil.
The alternative that usually gets used in productions with a moderate budget is a hardware camera with HDMI or SDI output into a Blackmagic capture device. There's still some latency, but it's better than a webcam. It's pretty much down to the minimum of what you can do with a computer -- the capture device still has to buffer a frame, transmit it to the computer, which copies it into RAM, then blits it to video RAM, where it waits for the next vertical refresh.
All that said, the human finger doesn't travel all that fast, so for a HID application like this, you could probably get the latency way down by extrapolating the near-future position of the finger, much like iPadOS does with Pencil.
[1] https://figure53.com/qlab