Method and system for automatic classification of objects

• US 8,063,815 B2
• Filed: 05/16/2006
• Issued: 11/22/2011
• Est. Priority Date: 05/15/2006
• Status: Expired due to Fees

• Current Assignee: Telefonaktiebolaget LM Ericsson
• Original Assignee: Telefonaktiebolaget LM Ericsson
• Inventors: Knapskog, Atle Onar, Valo, Hilde, Hjelmstad, Jens Fredrik
• Primary Examiner(s): Sotomayor; John
• Application Number:
  US12/300,972
• Publication Number:
  US 20090184865A1
• Time in Patent Office: 2,016 Days
• Field of Search: 342/25.R, 342 25 A- 25 F, 342/41, 342/64, 342/90, 342/179, 342/191, 342194-197
• US Class Current: 342/25.R
• CPC Class Codes: G01S 13/9064 Inverse SAR [ISAR]
• CPC Class Codes: G01S 7/412 based on a comparison between measured values and known or stored values
• CPC Class Codes: G06V 10/752 Contour matching
• CPC Class Codes: G06V 20/647 by matching two-dimensional images to three-dimensional objects

Abstract

A new method has been developed as an attempt to improve speed and robustness of existing ISAR classification methods. The new method produces a set of silhouettes of possible models in a 3D model database. The set of silhouettes of each model views the model from various viewing angles, as the target dimensions will vary as it is viewed from different angles. The silhouettes are stored as a training set. Classification is done by comparing the silhouette of the target with the set of silhouettes in the training set. The silhouettes are calculated prior to the silhouette matching.

Claims

1. A method of automatic classification of an observed object in Inverse Synthetic Aperture Radar (ISAR) image, comprising the steps of:

extracting a silhouette of the observed object,

extracting a length parameter for the silhouette of the observed object,

retrieving, based on the extracted length, training silhouettes for model objects having a similar length corrected for range from a database with a range of training silhouettes representing model objects of different classes and with different orientation angles,

comparing the silhouette of the observed object with said retrieved training silhouettes, and

classifying the observed object as being of the model object class with highest correlation between the silhouettes of the observed and model objects.

2. The method of claim 1, where the observed object is present in a sequence of images, said method including classifying the observed object in several images in parallel.

3. The method of claim 1, wherein the silhouettes are formed by an edge detection method.

4. The method of claim 1, said method including the additional step of selecting objects to be classified by determining a set of features describing the orientation of the observed object and selecting only those observed objects that are oriented so as to clearly display the shape of the object.

5. The method of claim 4, wherein said set of features includes a ship'"'"'s centreline, and determining the ship'"'"'s centreline using a Hough transform.

6. The method of claim 4, wherein said set of features includes a ship'"'"'s centreline, and determining the ship'"'"'s centreline by fitting a straight line through the ship'"'"'s hull using a least squares method.

7. The method of claim 4, wherein said set of features includes a ship'"'"'s hull side, and finding said hull side by fitting a second degree curve to a side of the observed object using a least squares method.

8. The method of claim 1, wherein the classification step is performed by finding best match between observed object silhouette and training silhouette using a Nearest neighbour classifier.

9. The method of claim 1, wherein said image is an Inverse Synthetic Aperture Radar image.

10. The method of claim 9, wherein the training silhouettes are extracted from Inverse Synthetic Aperture Radar Images.

11. The method of claim 1, wherein the training silhouettes are obtained from 3D wire-frame models.

12. The method of claim 1, wherein said image is an X-ray image, an optical image or an Infra Red optical image.

13. The method of claim 1, in which a classification result is presented to an operator as an object ID;

a code or a synthetic high definition optical image.

14. The method of claim 13, in which said classification result is presented together with information as to the importance of the object and/or a figure regarding the reliability of the classification result.

15. A system for automatic classification of an observed object in an Inverse Synthetic Aperture Radar (ISAR) image including a sensor supplying at least one image showing an observed object of interest, comprising:

a pre-processing unit for extracting a silhouette of said observed object and extracting a length parameter for the silhouette of the observed object,

a database with stored training silhouettes of classified model objects in several orientations, and

a classification unit for selecting, based on the extracted length, a set of training silhouettes of model objects having similar length parameters corrected for range as in the observed object and retrieve said selected set from the database,

said classification unit for comparing the silhouette of the observed object with said set of selected training silhouettes, and for classifying said observed object as being of the class corresponding to the training silhouettes which best matches the silhouette of the observed object.

16. The system of claim 15, where said sensor is an Inverse Synthetic Aperture Radar, or a X-ray sensor, or an optical sensor, or an Infra Red optical sensor.

17. The system of claim 15, where said sensor is a sea, ground or flying radar.

18. The system of claim 15, in which said preprocessing unit, database and classification unit are connected in a network.