Image analysis – Applications – Target tracking or detecting
Reexamination Certificate
1999-09-22
2003-07-22
Mehta, Bhavesh M. (Department: 2625)
Image analysis
Applications
Target tracking or detecting
C382S190000, C382S224000
Reexamination Certificate
active
06597800
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
This invention relates to an apparatus and a process for automatic recognition of a target object from an infrared or visible light image.
2. Discussion of Prior Art
Conventional Automatic Target Recognition systems involve capturing the image of a real field of view such as by an infrared camera utilising a two dimensional array of light intensity sensitive pixels or a video camera and carrying out processing on the image data to separate and identify objects appearing in the image. The objects are separated from the background and identified or classified to separate classes of object of interest to the viewer from objects of no interest. Conventionally an automatic target recognition system first segments the image to provide regions of interest that are homogeneous in some respect with the primary aim of extracting individual complete objects for later processing. Data about the objects produced in this segmentation are then passed to a feature extraction stage in which object features are extracted which have been chosen optimally to separate the different classes of object present. The features that have been extracted are then used to classify the objects that have been segmented.
This conventional approach suffers from the disadvantage that it is highly sensitive to the complexities of the real image data with the correct classification of target objects being highly dependent on the quality of the object segmentation. If the segmentation process does not produce object data containing features wich accurately represent the object class, then the objects themselves cannot be correctly classified and identified or identified at all.
There is thus a need for a generally improved apparatus and process for automatic recognition of a target object from an infrared or visible light image which improves the accuracy of target object recognition and identification.
SUMMARY OF THE INVENTION
According to one aspect of the present invention there is provided apparatus for automatic recognition of a target object from an infrared or visible light image, including an image producing device;
a primary separator for subjecting the image to primary segmentation in which the image is divided up into one or more primary homogeneous regions each approximating to an object of interest and data is extracted from the image about these primary regions,
a first feature extraction device for receiving the extracted primary region data from the primary separator and recognising and extracting features from the extracted primary region data, which features have been predetermined to separate objects in the primary regions into different classes,
a first classifying means for receiving the extracted features and classifying them thereby to recognise the or each object in the primary regions or to indicate that one or more of the objects is unclassified and therefore unrecognised,
a secondary segmentation unit for receiving from the primary separator data about the original segmented image primary region containing an unrecognised object and for submitting this data to secondary segmentation to provide sub regions of greater homogeneity, and for extracting data from the primary region about the sub regions,
a third feature extraction device for recognising and extracting classifying features from the extracted sub region data and,
a second classifying means for receiving the extracted sub-region classifying features and utilising them to classify and thereby recognise the or each object in the sub regions or to recognise that the or each object in the sub regions is not a target object.
Preferably the apparatus includes more than one secondary segmentation unit.
Conveniently the first classifying means includes a first classifier for receiving the extracted features from the first feature extraction device and sorting and classifying the extracted features, a first comparator for receiving, in parallel with the first classifier, the extracted features from the first feature extraction device, and generating a value of the probability that the classification by the first classifier is correct, and first assessment means for receiving data from the first classifier and first comparator and determining recognition or non recognition of the or each object in the primary regions.
Advantageously the second classifying means includes a second classifier for receiving the extracted sub region classifying features from the third feature extraction device and sorting and classifying the extracted sub region features, a second comparator, in parallel with the second classifier, for receiving the extracted sub region classifying features from the third feature extraction device, and generating a value of the probability that the classification by the second classifier is correct, and second assessment means for receiving data from the second classifier and second comparator and determining recognition or non recognition of the or each object in the sub regions.
Preferably the first and second assessment means are combined in a single assessor unit.
Conveniently the apparatus including a frame memory store for receiving and storing data from the image producing device for passage to the primary separator.
Advantageously the apparatus includes a feature store for receiving data from the first feature extraction unit and for passing the data to the first classifying means.
Preferably the apparatus includes a secondary store for receiving and storing image data from the frame memory store for passage to the secondary segmentation unit.
Conveniently the primary separator is operable to output bounding box data to the first feature extraction device, and including an image region extending device operable to receive the primary segmentation bounding box data outputted from the first feature extraction device extend the image region described by the bounding box data and pass the extended region data as a control signal to the secondary store and to the frame memory store.
According to another aspect of the present invention there is provided a process for automatic recognition of a target object from an infrared or visible light image, including the steps of
subjecting the image to primary segmentation in which the image is divided up into one or more primary homogeneous regions each approximating to an object of interest,
extracting data from the image about these primary regions,
recognising and extracting features from the extracted primary regions, which features have been predetermined to separate objects in the primary regions into different classes,
utilising the extracted features to classify and thereby recognise the or each object in the primary regions or to indicate that one or more of the objects is unclassified and therefore unrecognised,
subjecting the original segmented image primary region containing an unrecognisable object to secondary segmentation to provide sub regions of greater homogeneity,
extracting data from the primary region about the sub regions,
recognising and extracting classifying features from the extracted sub region data and,
utilising the extracted classifying features to classify and thereby recognise the or each object in the sub regions or to recognise that the or each object in the sub regions is not a target object.
Preferably the process includes more than one secondary segmentation.
Conveniently the data relating to the object recognised or unrecognised from the primary segmentation steps is compared with data relating to the object recognised or unrecognised from the or further secondary segmentation steps and assessed to provide object recognition or rejection.
Advantageously the assessment involves the production and consideration of a probability estimate value for each segmentation.
Preferably the primary segmentation is carried out by bounding the primary regions in dependence upon the intensity of illumination change at the boundary.
Conveniently the or each secondary segmentation is carried out by passing the
Murray Alan
Reavy Richard
Azarian Seyed
BAE Systems PLC
Mehta Bhavesh M.
Nixon & Vanderhye P.C.
LandOfFree
Automatic target recognition apparatus and process does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Automatic target recognition apparatus and process, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Automatic target recognition apparatus and process will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3070746