The integration of object motion information to pixel-wise foreground segmentation is investigated for moving object detection applications. It is achieved by using a motion feedback mask as a basis to adapt parameter of background subtraction based algorithm. This binary mask indicates the possible location of foreground pixel via forwarding the previously detected object location and size. All pixels are accordingly separated into two groups, and applied with distinct thresholding parameters when evaluating the difference to modeled background. Morphological smoothing and connected component labeling are conducted subsequent to segmentation stage to produce object-level foreground information. Look ahead matching of this object component information yields the object motion data. Experimental result shows this adaptive control is able to simultaneously reduce the disturbance of imager noise and camouflage effect which are artifacts produced by using parameters from opposite ends.
[1]
Monique Thonnat,et al.
Controlling background subtraction algorithms for robust object detection
,
2009,
ICDP.
[2]
W. Eric L. Grimson,et al.
Adaptive background mixture models for real-time tracking
,
1999,
Proceedings. 1999 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (Cat. No PR00149).
[3]
Chun-Jen Chen,et al.
A linear-time component-labeling algorithm using contour tracing technique
,
2004,
Comput. Vis. Image Underst..
[4]
Kenji Suzuki,et al.
Linear-time connected-component labeling based on sequential local operations
,
2003,
Comput. Vis. Image Underst..
[5]
Kesheng Wu,et al.
Two Strategies to Speed up Connected Component Labeling Algorithms
,
2005
.