Is it best to use opencv on its own or using opencv with trained model when detecting 2D signs through a live camera feed?

https://www.youtube.com/watch?v=Fchzk1lDt7Q

In this tutorial the person shows how to detect these signs etc without using a trained model.

However through a live camera feed I want to be able to detect these signs in real time. So which one would be better, to just use OpenCV on its own or to use OpenCV with a custom trained model such as pytorch etc?