System and method for object recognition based on three-dimensional adaptive feature detectors

• US 8,687,898 B2
• Filed: 02/01/2010
• Issued: 04/01/2014
• Est. Priority Date: 02/01/2010
• Status: Active Grant

• Current Assignee: Toyota Jidosha Kabushiki Kaisha (Toyota Motor Corporation)
• Original Assignee: Toyota Motor Engineering & Manufacturing North America Incorporated (Toyota Motor Corporation)
• Inventors: Prokhorov, Danil V., James, Michael R.
• Primary Examiner(s): Repko, Jason M
• Assistant Examiner(s): ROWLAND, JHERE M
• Application Number:
  US12/697,808
• Publication Number:
  US 20110188737A1
• Time in Patent Office: 1,520 Days
• Field of Search: None
• US Class Current: 382/224
• CPC Class Codes: G06F 18/00 Pattern recognition

Abstract

Method and system for imaging an object in three-dimensions, binning data of the imaged object into three dimensional bins, determining a density value p of the data in each bin, and creating receptive fields of three dimensional feature maps, including processing elements O, each processing element O of a same feature map having a same adjustable parameter, weight Wc1. The density values p are processed with the processing elements O to determine an output value o for at least each three dimensional location of the imaging of the object having a density value p above a predetermined threshold value. An object is classified by processing the output values o with a committee of classifiers, and the classification of the object is communicated.

Claims

1. A method, comprising:

imaging an object in three dimensions by collecting three-dimensional data points of the object;

binning the data points of the imaged object into three dimensional bins having a predetermined three-dimensional size to create a binned scanned object in a three-dimensional space;

determining a density value of the data in each of the bins by calculating a number of the data points in each of the bins;

creating receptive fields of three dimensional feature maps, including processing elements, each of the processing elements of a same feature map having a same adjustable parameter, which is defined as a weight;

generating an output value, by processing the density values with the processing elements, for at least each three dimensional location of the imaging of the object having a density value above a predetermined threshold value;

classifying the object by processing the output values with a committee of classifiers, which includes a cascading series of classifiers; and

communicating the classification of the object.

2. The method according to claim 1, wherein the determining the density value of the data in each bin is based on a viewing angle of the object, which is defined as an index.

3. The method according to claim 2, further comprising:

normalizing the density values separately for each index.

4. The method according to claim 1, wherein the cascading series of classifiers are weak classifiers, which are trained using an AdaBoost process.

5. The method according to claim 4, further comprising:

ranking the output values generated by processing the density values p with the processing elements, selecting maximum values of the output values, and recording a location of the maximum values of the output values in the respective feature map, and

at least one of the weak classifiers is trained using the AdaBoost process with the selected maximum values of the output values and the recorded locations of the maximum values.

16. The method according to claim 1, wherein the processing elements are transformations that are applied to a product of the weight and the density values in the processing the density values with the processing elements.

17. The method according to claim 16, wherein the processing elements are nonlinear transformations.

6. A non-transitory computer readable medium including computer executable instructions, which when executed by a processor, cause a computer system to perform a method, the method comprising:

imaging an object in three dimensions by collecting three-dimensional data points of the object;

binning the data points of the imaged object into three dimensional bins having a predetermined three-dimensional size to create a binned scanned object in a three-dimensional space;

determining a density value of the data in each of the bins by calculating a number of the data points in each of the bins;

creating receptive fields of three dimensional feature maps, including processing elements, each of the processing elements of a same feature map having a same adjustable parameter, which is defined as a weight;

generating an output value, by processing the density values with the processing elements, for at least each three dimensional location of the imaging of the object having a density value above a predetermined threshold value;

classifying the object by processing the output values with a committee of classifiers, which includes a cascading series of classifiers; and

communicating the classification of the object.

7. The non-transitory computer readable medium according to claim 6, wherein the determining the density value of the data in each bin is based on a viewing angle of the object, which is defined as an index.

8. The non-transitory computer readable medium according to claim 7, the method further comprising:

normalizing the density values separately for each index.

9. The non-transitory computer readable medium according to claim 6, wherein the cascading series of classifiers are weak classifiers, which are trained using an AdaBoost process.

10. The non-transitory computer readable medium according to claim 9, the method further comprising:

ranking the output values generated by processing the density values p with the processing elements, selecting maximum values of the output values, and recording a location of the maximum values of the output values in the respective feature map, and

at least one of the weak classifiers is trained using the AdaBoost process with the selected maximum values of the output values and the recorded locations of the maximum values.

18. The non-transitory computer readable medium according to claim 6, wherein the processing elements are transformations that are applied to a product of the weight and the density values in the processing the density values with the processing elements.

19. The non-transitory computer readable medium according to claim 18, wherein the processing elements are nonlinear transformations.

11. A computer system, comprising:

means for imaging an object in three dimensions by collecting three-dimensional data points of the object;

means for binning the data points of the imaged object into three dimensional bins having a predetermined three-dimensional size to create a binned scanned object in a three-dimensional space;

means for determining a density value of the data in each of the bins by calculating a number of the data points in each of the bins;

means for creating receptive fields of three dimensional feature maps, including processing elements, each of the processing elements of a same feature map having a same adjustable parameter, which is defined as a weight;

means for generating an output value, by processing the density values with the processing elements, for at least each three dimensional location of the imaging of the object having a density value above a predetermined threshold value;

means for classifying the object by processing the output values with a committee of classifiers, which includes a cascading series of classifiers; and

means for communicating the classification of the object.

12. The computer system according to claim 11, wherein the means for determining determines the density value of the data in each bin based on a viewing angle of the object, which is defined as an index.

13. The computer system according to claim 12, further comprising:

means for normalizing the density values separately for each index.

14. The computer system according to claim 11, wherein the cascading series of classifiers are weak classifiers, which are trained using an AdaBoost process.

15. The computer system according to claim 14, wherein

the means for generating the output value by processing the density values with the processing elements ranks the output values, selects maximum values of the output values, and records a location of the maximum values of the output values in the respective feature map, and

at least one of the weak classifiers is trained using the AdaBoost process with the selected maximum values of the output values and the recorded locations of the maximum values.

20. The computer system according to claim 11, wherein the processing elements are transformations that are applied to a product of the weight and the density values in the processing the density values with the processing elements.

21. The computer system according to claim 20, wherein the processing elements are nonlinear transformations.