Clutter rejection using adaptive estimation of clutter probability density function

Claim 2

A signal processor for identifying a target contained in an image in the presence of interference, said signal processor comprising: a target model for processing target feature estimates; a detection processor for processing input image data comprising the image and for generating regions of interest and features contained in the image; a temporal processor for processing the regions of interest and features contained in the image output by the detection processor, and for generating tracks and features contained in the image; and a target processor coupled to the target model and the temporal processor for processing the tracks and features contained in the image and the target feature estimates to adaptively estimate the feature probability density function of the interference from the image data, and for processing the estimated probability density function of the interference and the target feature estimates using a Bayesian classifier to generate a signal indicative of the target contained in the image in the presence of interference.

Claim 3

The signal processor of claim 1 wherein the input image is provided by an imager.

Claim 4

The signal processor of claim 1 wherein the target processor comprises means for: computing the probability density function of the target using the expected target feature vector and the covariance; for each track, computing the aposteriori probability that the track belongs to the target class; computing a normalized square distance between all track pairs of confirmed tracks derived from the temporal processor; estimating the feature vector and the feature covariance of a nontarget class using the sampled mean and variance of the feature vectors of the members of the nontarget class; estimating a feature probability density function of the nontarget class using the class feature and covariance; for each confirmed track, computing the aposteriori probability that the track belongs to of the nontarget class; computing a likelihood ratio for each confirmed track; finding all candidate tracks whose feature vectors are greater than a first predetermined threshold and whose likelihood ratio are greater than a second threshold; selecting the track with the highest likelihood ratio which is the best target track of the image; and acquiring a track that is persistently the best track over several consecutive images as the target.

Claim 5

The signal processor of claim 3 wherein tracks are clustered on the basis of normalized square distance using a hierarchical agglomeration process.

Claim 6

The signal processor of claim 3 wherein the nontarget class comprises each cluster having three or more members.

Claim 7

The signal processor 1 of claim 3 wherein the feature probability density function of the nontarget class is estimated using the class feature and covariance assuming a multivariate normal distribution.

Claim 10

The method claim 7 wherein the step of adaptively estimating the feature probability density function of the interference comprises the steps of: computing the probability density function of the target using the expected target feature vector and the covariance; for each track, computing the probability that the track belongs to the target class; computing normalized square distance between all track pairs of confirmed tracks derived from the temporal processor; estimating 3 the feature vector and the feature covariance of a nontarget class using the sampled mean and variance of the feature vectors of the members of the nontarget class; estimating a feature probability density function of the nontarget class using the class feature and covariance; for each confirmed track, computing unnormalized probability that the track belongs to the nontarget class; and computing a likelihood ratio for each confirmed track.

Claim 11

The method claim 7 wherein the step of processing the estimated interference probability density function and the target feature estimates using a Bayesian classifier comprises the steps of: finding all candidate tracks whose feature vectors are greater than a first predetermined threshold and whose likelihood ratio are greater than a second threshold; selecting the track with the highest likelihood ratio which is the best target track of the image; and acquiring a track that is persistently the best track over several consecutive images as the target.

Claim 12

The method claim 8 further comprising the steps of: finding all candidate tracks whose feature vectors are greater than a first predetermined threshold and whose likelihood ratio are greater than a second threshold; selecting the track with the highest likelihood ratio which is the best target track of the image; and acquiring a track that is persistently the best track over several consecutive images as the target.