Virtual reality scene generator for generating training images for a pattern recognition classifier

• US 7,574,018 B2
• Filed: 12/07/2005
• Issued: 08/11/2009
• Est. Priority Date: 12/07/2005
• Status: Expired due to Fees

• Current Assignee: TRW Automotive USLLC (Zeppelin - Stiftung der Stadt Friedrichshafen)
• Original Assignee: TRW Automotive USLLC (Zeppelin - Stiftung der Stadt Friedrichshafen)
• Inventors: Luo, Yun
• Primary Examiner(s): BHATNAGAR, ANAND P
• Application Number:
  US11/299,129
• Publication Number:
  US 20070127811A1
• Time in Patent Office: 1,343 Days
• Field of Search: 382/100, 382/103, 382/104, 382/106, 382/107, 382/181, 382/190, 382/195, 701/1, 701/28, 701/45
• US Class Current: 382/103
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06V 10/772 Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries
• CPC Class Codes: G06V 40/103 Static body considered as a whole, e.g. static pedestrian or occupant recognition

Abstract

Systems and methods are provided for generating training samples for a classification system associated with a vehicle occupant protection system. A three-dimensional virtual reality scene is generated (138) from a cockpit model having a set of at least two user configurable parameters (132) and an occupant model having a set of at least two user configurable parameters and a plurality of points of articulation (134). A plurality of training samples are generated from the virtual model (142). Each of the plurality of training samples has a first set of parameters associated with the cockpit model, a second set of parameters associated with the occupant model, and an occupant pose defined according to the points of articulation associated with the occupant model.

Claims

1. A computer readable medium, encoded with executable instructions configured to be executed by a computer processor to perform a method for generating training samples for a classification system associated with a vehicle occupant protection system, the method comprising the following steps:

generating a three-dimensional virtual reality scene from a cockpit model having a set of at least two user configurable parameters, and an occupant model having a set of at least two user configurable parameters and a plurality of points of articulation; and

generating a plurality of samples from the virtual reality scene, each of the plurality of training samples having a first set of parameters associated with the cockpit model, a second set of parameters associated with the occupant model, and an occupant pose defined according to the points of articulation associated with the occupant model.

2. The computer readable medium of claim 1, the method further comprising the step of generating a plurality of verification samples from the virtual scene, each of the plurality of verification samples having a first set of parameters associated with the cockpit model, a second set of parameters associated with the occupant model, and an occupant pose defined according to the points of articulation associated with the occupant model.

3. The computer readable medium of claim 2, the method further comprising the steps of:

training a pattern recognition classifier on the plurality of training samples;

classifying the plurality of verification samples at the pattern recognition classifier; and

determining a fitness metric for the pattern recognition classifier as a function of at least one of an accuracy and a rejection rate associated with the classifier according to known class memberships of the plurality of verification samples.

4. The computer readable medium of claim 3, the method further comprising the steps of:

identifying a set of at least one of the plurality of verification samples that was not correctly classified by the pattern recognition classifier; and

training the pattern recognition classifier on the identified set of verification samples if the fitness metric for the pattern recognition classifier does not meet a threshold value.

5. The computer readable medium of claim 3, the method further comprising the steps of:

taking a sensor reading of the vehicle interior at a sensor located within the vehicle;

classifying the sensor reading at the pattern recognition classifier into a selected one of a plurality of output classes; and

controlling the vehicle occupant protection system according to the selected output class.

6. The computer readable medium of claim 1, wherein the step of generating a three-dimensional virtual reality vehicle scene comprises the steps of:

assigning values for the set of at least two parameters associated with the cockpit model as to generate an instance of the cockpit model having at least one desired characteristic;

assigning values for the set of at least two parameters associated with the occupant model as to generate an instance of the occupant model having at least one desirable characteristic; and

fusing the generated instance of the cockpit model and the generated instance of the occupant model to create the three-dimensional virtual reality scene.

7. The computer readable medium of claim 1, wherein the step of generating a plurality of training samples from the virtual scene includes generating a color image, a grayscale image, and a stereo disparity image from at least one camera perspective.

8. The computer readable medium of claim 7, further comprising the steps of:

extracting a first set of numerical features from the color image;

extracting a second set of numerical features from the grayscale image;

extracting a third set of numerical features from the stereo disparity image; and

constructing a composite feature vector for training a pattern recognition classifier from the first set of numerical features, the second set of numerical features, and the third set of numerical features.

9. The computer readable medium of claim 1, wherein the step of generating a plurality of training samples comprises at least one step from a step of generating a sequence of images in time and a step of simulating an output from an array of weight sensors.

10. A system for generating a plurality of samples, wherein the samples can be utilized as training and verification samples for a pattern recognition classifier, comprising:

a memory storing;

a cockpit model, configurable according to a plurality of parameters, that simulates a vehicle interior in a three-dimensional space

an occupant model, configurable according to a plurality of parameters, that simulates a vehicle occupant;

a virtual reality scene constructor that fuses the occupant model and the cockpit model to form a three-dimensional virtual reality scene; and

an image generator that generates an image of the three-dimensional virtual reality scene from a selected camera perspective; and

a processor, operatively connected to the memory such that the processor can access data stored on the memory, the processor being configured to access each of the cockpit model, the occupant model, the virtual reality scene constructor, and the image generator as to generate the plurality of samples.

11. The system of claim 10, wherein the cockpit model includes a seat model that simulates a vehicle seat according to a plurality of parameters, at least one of the plurality of parameters associated with the seat model defining the position of the seat within the cockpit.

12. The system of claim 11, wherein the plurality of parameters associated with the seat model includes at least one parameter defining the orientation of the seat along at least one range of motion.

13. The system of claim 10, wherein the plurality of parameters associated with the occupant model includes at least one parameter that defines a pose of the occupant along at least one point of articulation.

14. The system of claim 10, wherein the image generator is operative to generate a plurality of images of the virtual scene, where the associated camera perspectives of the plurality of images are selected such that a stereo disparity map can be generated from the plurality of images.

15. The system of claim 10, wherein the system comprises a plurality of occupant models, each representing an associated occupant type, and one of the plurality of occupant models is selected for use in the virtual reality scene.

16. The system of claim 15, wherein the selected occupant model represents an adult human, and at least one of the plurality of parameters associated with the occupant model define a height and weight of the occupant.

17. The system of claim 10, wherein the plurality of parameters associated with the occupant model includes at least one parameter that defines at least one chromatic property of the surface of the occupant.

18. A computer program product, encoded on a computer readable medium and operative in a data processing system, for determining at least one characteristic of a vehicle occupant, comprising:

a component for generating a set of verification samples, having a known class membership, for the pattern recognition classifier, comprising;

a cockpit model, configurable according to a plurality of parameters, that simulates a vehicle interior in a three-dimensional space;

an occupant model, configurable according to a plurality of parameters, that simulates a vehicle occupant; and

a sample generator that generates a sample for the classifier from the cockpit model and the occupant model; and

a pattern recognition classifier operative to classify the set of verification samples, identify a set of at least one verification sample that was not properly classified from the set of verification samples, and incorporate the identified set of at least one verification sample into a set of training data associated with the classifier.

19. The computer program product of claim 18, wherein the component for generating a set of verification samples is operative to generate a new set of verification samples for the pattern recognition classifier when a fitness parameter associated with the pattern recognition classifier does not meet a threshold value.

20. The computer program product of claim 18, wherein the plurality of parameters associated with the occupant model includes at least one parameter that defines the pose of the occupant along at least one point of articulation.

21. The computer program product of claim 18, wherein the component for generating a set of verification samples comprises a virtual reality scene constructor that fuses the occupant model and the cockpit model to form a three-dimensional virtual reality scene, the image generator being operative to generate at least one two-dimensional representation of the three-dimensional virtual reality scene according to an associated camera perspective.