Skip to main content

Gesture Recognition

Same chip (ndp120_b0) and architecture options as Audio Classification. Architecture selection depends on the complexity of the gestures and sensor input.

Chip and Architecture

Same chip (ndp120_b0) and architecture options as Audio Classification. Architecture selection depends on the complexity of the gestures and sensor input.

Time-Series Feature Configuration — User Inputs

ParameterDescription
Input Matrix (Features)Values per frame — typically equals sensor channels or derived features per time step
Input Matrix (Num Frames)Number of time frames stacked along the time axis
Window Duration (T)Total observation window duration in seconds
ChannelsNumber of independent sensor channels (e.g. 6 for 3-axis accel + 3-axis gyro)
Window Step (Hop Size)Samples to advance between successive windows
Window Length (Samples)Raw samples in each individual short-time window

Example Parameters — 6-Channel IMU (ndp120_b0)

ParameterValue
Input Matrix (Features)36
Input Matrix (Num Frames)40
Window Duration (T)2.40 s
Channels6
Window Step (Hop Size)6
Num Samples to NN (N, calculated)1440
Calculated Sampling Rate100.00
NOTE

For 6-DoF IMU gesture recognition, a 2.4 s window at 100 Hz is a robust starting point. Adjust Window Duration and Channels to match your sensor hardware.

Label Selection

FieldDescription
Target GesturesGestures the model should detect — each is an individual class
Open-set WordsNon-target inputs merged into one class
Number of ClassesAuto-calculated: target gestures + 1 — read-only