PUBLIC

FeatureTrackC::FeatureTrackC(UIntT) FeatureTrackC::Filter(DListC &) const

FeatureTrackC
Performs history analysis of the tracked objects, ie temporal filtering
include	"amma/DERA/FeatureTrack.hh"
User Level:	Default
Library:	DERARegTrack
Example:	exIndex.cc
Section:	Applications.Image Processing.DERA Small Target Detection
In Scope:	std

Comments:
FeatureTrackC is responsible for performing temporal filtering for the DERA Small Target Tracking project. The purpose is to invesigate the history of feature measurements to determine whether the target is interesting or not.

The algorithm has impact in 3 main ways:

1. Each target to be considered must have been successfully tracked for at least n consecutive frames, where n is the length of history specified in the constructor.
2. Linear models are fitted to region size, average and gradient contrast, and compactness using the measured values in the history and their frame index as ordinate value. Tests are then performed on each feature to determine if the feature is behaving predictably by calculating the ratio of the predicted feature value for the current frame to the actual value for the current frame. The different features have to satisfy the following restrictions:

    0.6 < size ratio < 1.66
    0.8 < average contrast ratio < 1.25
    0.8 < gradient contrast ratio < 1.25
    0.8 < compactness ratio < 1.25
    

3. Very small stationary targets are eliminated by finding the average motion of the target during the history. The target is rejected if the mean motion magnitude is less than 0.3 pixels and the target is less than 4 pixels in size.

FeatureTrackC(UIntT history)

Constructor

ind	Index of types of intererest.

DListC<RegBoundaryC> Filter(DListC<RegBoundaryC> & before) const

Prunes list of regions according to their history

ind	Index of types of intererest.