hello, please tell me how to correctly type data for gestures, I tried to type continuous movement, it doesn’t work out very well, since the gesture is not done all the time
Hello @norik.badalyan,
I am not sure I completely understood your use case?
Do you want to detect single gesture?
If so, you can split your samples into individual gestures and you will need to add plenty of noise data (movements where you are not doing this gesture). You can call it noise or unknown.
In that case, when you are not doing your gesture, it will be grabbed by this “noise / unknown” class.
Regards,
Louis
1 Like
I am collecting data for gesture communication, but I am not sure if I am collecting data correctly, please tell me how to collect data for this.
this is how I collected data when the gesture transition classification starts to make prediction of the same class two or more times
prediction gives the same multiple times how to fix it??
when the board is stationary it offers class 0, then when I make one move it offers class 1 two or more times.
Hello @norik.badalyan,
Are you running the C++ library? Arduino library? Or the standalone firmware?
In the provided firmwares, we grab the data and the run the inference over and over.
The libraries, examples and firmwares are provided so you can start from here and then modify the code, all are open source.
In most use cases, users want to modify it to apply a custom business logic.
Can you explain what you are trying to achieve in details, how did you collect your data, which board and sensors do you use? I see that you are using 4 accelerometers right?
Regards,
Louis
Hello Louis,
I am running the C++ library. My board is based on a stm32 microcontroller and 5 accelerometers. I am trying to train model for recognizing dumb people’s gestures. 5 accelerometres are for 5 fingers respectively․
Regards,
Norik
Hello @norik.badalyan,
Ok great, then in your C++ code, you’ll need to add your custom logic to decide when to collect the accelerometer data, put that in a buffer and run the inference at the appropriate moment and then stop the inference when it suits you.
Or you can also filter out the noise class, apply an average on the prediction to only print one result and pass that result to some other functions in your code.
Regards,
Louis
1 Like