Multiple measurements are recorded for each gesture. Test code. Function call.
The measures captured here are acceleration in different directions:
| ax (m/s^2) | ay (m/s^2) | az (m/s^2) | aT (m/s^2) |
Looking at a single measure like x, we can see a single measure may not differentiate gestures. The multi measure plots above look much more different for different gestures. Test code. Function call.
A single measure may not be very different between classes so it’s important to have enough combinations of measures to distinguish classes. Visualizing data is very import to check your data. If you cannot see any differences, a trained model would have more difficulty and need more measures. At larger number of measures it may become more difficult to differentiate by eye. For training models with multiple measures they will need to be bundled into tensors.
A single motion we want to detect covers a window of the continuous measurements. We want to break a continuous recordings of measures of different gestures into windows roughly covering a single motion that we can distinguish within the window. Plotting the data and trying windows will help us pick window to test though we may try different window sizes to training a model to find the window that distinguishes the data the best. Test code. Function call.
Plotting the data and looking at window sizes can be used to select an initial window size as a starting point. We may try different window sizes during model training later to select the best window size. After selecting a window size that roughly covers full gestures for the different gestures and data will need to be processed into blocks of tensors containing all measurements for the windows in a later post.
Learn IoT and AI 