I am unsure about which approach is "better" (less time good accuracy) for classifying gestures:

The system is a doppler radar that returns I and Q signals

Approach 1: Have several subject perform gestures and label all gestures Then train a ML model (KNN, SVM) cross-validate etc... This approach requires that I collect a lot of data from several subjects

Approach 2: Use peak analysis, finding the number of peaks that represent a gesture I also use the distance between peaks as a characteristic of a gesture This requires no training data but rather looking at the raw data determining peaks and distances between peaks and mapping that to some criteria for a gesture This saves a lot of time and I don't have to collect so much data from subjects

It seems as though the solution is always to use machine learning, but I could easily do this by observing peaks.

How can I determine what approach to use?

  • $\begingroup$ implement both, and evaluate performance? $\endgroup$ – user28715 Feb 11 '19 at 16:18
  • 2
    $\begingroup$ as a general heuristic, simple is typically better and should be tried first. if not good enough, try the more complex. Either way, you should figure out your requirements before choosing an approach $\endgroup$ – user28715 Feb 11 '19 at 16:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.