gkthiruvathukal
commented
4 years ago Another option is to flag suspicious values. Once your numbers are above 80% accuracy, there are methods for identifying outliers and/or addressing missing data.
One thing you could do with your your argparse is to set thresholds. Max speed, max lat/long drift, etc. Given that the nominal use case is on a road trip, you're unlikely to be in Iowa during one reading and in West Virginin in the next. Furthermore, you are even unlikely to see a certain drive by calculating d(p1,p2) and assuming a max velocity. So if the OCR is bad, it is likely that the Euclidean distance needs to jibe with common sense.
This could include binarization, negation, sharpening, etc.