segmentation
Segment a video stream using motion information.
Model the contents of a field as a Guassian distribution.
Model the contents of a masked field as a Guassian distribution.
Model the contents of a masked field as a Guassian distribution.
This algorithm functions identically to GaussianSurfaceModel, but with additional support for masks. Rather than computing the mean and covariance of the entire field, MaskedGaussianSurfaceModel computes these parameters using only information in the masked region. Probability density is still estimated for the entire field.
Model the contents of a field as a Guassian distribution.
GaussianSurfaceModel computes the mean and covariance of the input, optionally smoothing the model over time, and uses these values to evaluate the probability density function over the field. High values in the output indicate points that are highly consistent with the Guassian model. Values close to zero indicate inconsistent points. This tends to pop out arbitrary unusual objects on approximately homogeneous backgrounds.
The temporal smoothing parameter controls the rate at which the model is permitted to change. With a value of 0, no smoothing is performed and a new model is computed for every frame. Larger values mix a percentage of the model from the previous frame. The smoothing parameter must be less than 1.
GaussianSurfaceModel supports both univariate and multivariate Gaussian models. ScalarField input will result in a one-dimensional model. VectorField input will produce a multivariate model. Note that the multivariate probability density function is approximated using an LSQR solver. This is more efficient, but also inexact. The multivariate version of GaussianSurfaceModel is dramatically less efficient than the univariate version. Use the univariate version whenever possible.