Method and apparatus for classifying a vehicle occupant via a non-parametric learning algorithm

• US 20070127824A1
• Filed: 12/07/2005
• Published: 06/07/2007
• Est. Priority Date: 12/07/2005
• Status: Abandoned Application

• Current Assignee: TRW Automotive USLLC (Zeppelin - Stiftung der Stadt Friedrichshafen)
• Original Assignee: TRW Automotive USLLC (Zeppelin - Stiftung der Stadt Friedrichshafen)
• Inventors: Luo, Yun, Su, Xiuling
• Application Number:
  US11/299,113
• Publication Number:
  US 20070127824A1
• Time in Patent Office: Days
• US Class Current: 382/224
• 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 classifying an input feature vector, representing a vehicle occupant, into one of a plurality of occupant classes. A database (106) contains a plurality of feature vectors in a multidimensional feature space. Each feature vector has an associated class from the plurality of output classes. A data pruner (108) eliminates redundant feature vectors from the database. A data modeler (109) constructs an instance-based, non-parametric classification model (110) in the multidimensional feature space from the plurality of feature vectors. A class discriminator (112) selects an occupant class from the plurality of occupant classes according to the constructed classification model.

Claims

1. A classification system for classifying an input feature vector, representing a vehicle occupant, into one of a plurality of occupant classes, comprising:

a database containing a plurality of feature vectors in a multidimensional feature space, each feature vector having an associated class from the plurality of output classes;

a data pruner that eliminates redundant feature vectors from the database;

a data modeler that constructs an instance-based, non-parametric classification model in the multidimensional feature space from the plurality of feature vectors; and

a class discriminator that selects an occupant class from the plurality of occupant classes according to the constructed classification model.

2. The system of claim 1, the data modeler being operative to construct a generalized binary search tree.

3. The system of claim 2, the data modeler being operative to construct the generalized binary search tree according to an approximate nearest neighbor algorithm.

4. The system of claim 3, the approximate nearest neighbor algorithm comprising a best bin first algorithm.

5. The system of claim 1, the classification model comprising a global density function and the data modeler being operative to define kernel functions around each of the feature vectors and construct the global density function from the defined kernel functions.

6. The system of claim 5, the data modeler being operative to compute a normalized sum of the kernel functions to produce the global density function.

7. The system of claim 1, further comprising a sensor assembly that monitors a vehicle interior to provide data associated with a vehicle occupant and a feature exactor that provides the input feature vector from the data associated with the vehicle occupant.

8. The system of claim 7, where the sensor assembly comprises an array of weight sensors located in a vehicle seat.

9. The system of claim 1, where the data pruner eliminates a feature vector when the feature vector falls within a first threshold distance of another feature vector having the same associated class and beyond a second threshold distance of all feature vectors having a different associated class.

10. The system of claim 1, the plurality of occupant classes comprising a class representing rearward facing infant seats.

11. A method for classifying an occupant into one of a plurality of output classes, comprising:

generating training data comprising a plurality of feature vectors in a multidimensional feature space, each feature vector having an associated class from the plurality of output classes;

eliminating redundant feature vectors from the training data, such that a feature vector from the plurality of feature vectors is eliminated when the feature vector falls within a first threshold distance in the multidimensional feature space of another feature vector having the same associated class and beyond a second threshold distance of all feature vectors having a different associated class;

constructing an instance-based, non-parametric classification model in the multidimensional feature space from the plurality of feature vectors;

extracting features from sensor data associated with a vehicle occupant, such that an input feature vector can be determined in the multidimensional feature space to represent the vehicle occupant; and

assigning an output class to the vehicle occupant according to the determined input feature vector and the constructed classification model.

12. The method of claim 11, wherein the step of constructing a classification model includes constructing a generalized binary search tree according to an approximate nearest neighbor algorithm.

13. The method of claim 12, the approximate nearest neighbor algorithm comprising a locality sensitive hashing algorithm.

14. The method of claim 11, wherein the classification model comprises a global density function and the step of constructing a classification model includes defining kernel functions around each of the feature vectors and constructing the global density function from the defined kernel functions.

15. The method of claim 11, the plurality of output classes comprising a class representing adult occupants of a vehicle.

16. The method of claim 11, further comprising the step of providing the assigned output class to a vehicle occupant protection system.

17. A computer program product, operative in a data processing system and embedded in a computer readable medium, for classifying a vehicle occupant comprising:

a database containing a plurality of feature vectors in a multidimensional feature space, each feature vector having an associated class from the plurality of output classes;

a data pruning module that eliminates redundant feature vectors from the plurality of feature vectors;

a data modeling module that constructs an instance-based, non-parametric classification model in the multidimensional feature space from the plurality of feature vectors; and

a class discriminator module that selects an occupant class for the vehicle occupant from the plurality of occupant classes according to the constructed classification model and an input feature vector representing the occupant.

18. The computer program product of claim 17, the data modeling algorithm being operative to construct a generalized binary search tree according to an approximate nearest neighbor algorithm.

19. The computer program product of claim 17, the classification model comprising a global density function and the data modeling algorithm being operative to define kernel functions around each of the feature vectors and construct the global density function from the defined kernel functions.

20. The computer program product of claim 17, the plurality of occupant classes comprising a class representing frontward facing child seats.