Video foreground segmentation method
A fully automatic, computationally efficient segmentation method of video employing sequential clustering of sparse image features. Both edge and corner features of a video scene are employed to capture an outline of foreground objects and the feature clustering is built on motion models which work on any type of object and moving/static camera in which two motion layers are assumed due to camera and/or foreground and the depth difference between the foreground and background. Sequential linear regression is applied to the sequences and the instantaneous replacements of image features in order to compute affine motion parameters for foreground and background layers and consider temporal smoothness simultaneously. The Foreground layer is then extracted based upon sparse feature clustering which is time efficient and refined incrementally using Kalman filtering.
1. For a video image including both a foreground layer and a background layer a method of segmenting the foreground layer from the background layer, said method comprising the computer implemented steps of:
extracting sparse features from a series of image frames thereby producing a sparse feature set for each of the individual images in the series;
performing a sequential linear regression on the sparse feature sets thereby producing a sequential feature clustering set;
extracting the foreground layer from the background layer using the sequential feature clustering set;
refining the extracted layer;
determining optical flows of the sparse features between consecutive frames;
determining a set of features including both edge features and corner features;
computing a covariance matrix for each individual feature to determine if the feature is an edge or a corner feature, wherein a covariance matrix is computed for each individual feature to determine if the feature is an edge or a corner feature;
computing, for each edge feature, its normal direction (dx, dy) from the covariance matrix; and
projecting its optical flow to this nominal direction.
2. The method according to claim 1 wherein a set of features and their optical flow values between two frames is defined by: (δx i ,δy i ),i=1, . . . , n where n is the number of features, said method further comprising the steps of:
comparing two sets of affine parameters, and
classifying features to each set.
3. The method of claim 2 wherein said comparing and classifying steps further comprise the steps of:
randomly clustering the features into two sets;
determining least square solutions of the affine parameters for each set of features, and use normal optical flow for edge features;
fitting each feature into both affine motion models and comparing residuals;
classifying each feature to the affine model depending upon the residual;
repeating, the determining, fitting and classifying steps above until the clustering process converge.
4. The method of claim 3 further comprising the step of:
extending the feature clustering from two frames to several frames.