@smoothing attribute

signature

smoothing SMOOTH_FLOAT1 [SMOOTH_FLOAT2 SMOOTH_FLOAT3 SMOOTH_FLOAT4 SMOOTH_FLOAT5]

values

sensor specific default

examples

@smoothing 0.5 0.1 0.5 0.1 0.1
@smoothing 0.1

Smoothing for skeleton joint values with either a simplified one parameter factor or five parameter Holt double-exponential filter.

1-parameter: simplified temporal smoothing factor where 0.0 is no smoothing and 1.0 is extreme smoothing. Less smoothing will decrease the lag of joints yet cause more position and orientation jitters.
5-parameter: smoothing based on a Holt double-exponential filter with the five Holt parameters of smoothing, correction, prediction, jitter-reduction radius, and max-deviation radius

Holt double-exponential parameters

When a plugin uses Holt, the default smoothing parameters are tuned for gesture recognition. These default values may not be best for your needs.

smoothing: Range is [0.0..1.0] and the default value is 0.5. A higher value corresponds to more smoothing; passing 0 causes the raw data to be returned. As the smoothing value is increased, responsiveness to the raw data decreases; therefore, increased smoothing leads to increased latency.
correction: Range is [0.0..1.0] and the default value is 0.5. A higher value corrects toward the raw data more quickly. Too high a value can make values “springy”. A lower value corrects more slowly and appears smoother.
prediction: Range is [0.0..inf] and the default value is 0.5 frames. This is the number of frames to predict into the future. A value greater than 0.5 will likely lead to overshoot when the data changes quickly; dampen the overshoot effect by using a smaller value of max-deviation radius.
jitter-reduction radius: Range is [0.0001..inf] and the default value is 0.05 meters. This is the radius (meters) that defines a jitter value and therefore how that value is smoothed. Values outside this radius are considered new positions and smoothed differently.
max-deviation radius: Range is [0.0..inf] and the default value is 0.04 meters. It is the maximum radius (meters) that a filtered position can deviate from raw data. A filtered value that exceeds this radius is clamped at the maximum radius in the direction of the filtered value. Too high a value can snap back to noisy data.

Microsoft’s documentation and smoothing whitepaper provide suggestions for the 5-parameter Holt filter.

0.5 0.5 0.5 0.05 0.04 some smoothing with little latency; only filters out small jitters, good for gesture recognition (this is the default)
0.5 0.1 0.5 0.1 0.1 smoothed with some latency; filters out medium jitters, good for a menu system that needs to be smooth but doesn’t need the reduced latency as much as gesture recognition
0.7 0.3 1.0 1.0 1.0 very smooth and much latency; filters out large jitters, good for situations where smooth data is absolutely required and latency is not an issue