You own a vision pipeline used in a head worn AR device. The system works in production, but accuracy is inconsistent across lighting, motion, users, and environments, and product teams want a clear plan to improve it.
If tasked with improving the accuracy of a vision system, what steps would you take?