YOLO and opencv really show how far we have come in the past decade.

I think one place that really needs work is hardware. Sensors are limited in frequency range, and refresh rate or resolution. And a big bottle neck for computer vision is that you need a large bandwidth provider if you are doing centralized processing of the feeds and not on edge processing.