Vehicle seat occupant classifying method and apparatus based on a support vector machine

• US 20030218321A1
• Filed: 05/23/2003
• Published: 11/27/2003
• Est. Priority Date: 05/24/2002
• Status: Active Grant

• Current Assignee: DENSO Corporation
• Original Assignee: DENSO Corporation
• Inventors: Tamura, Shinichi, Suzuki, Tomoji
• Granted Patent: US 6,889,129 B2
• Time in Patent Office: Days
• US Class Current: 280/730.1
• CPC Class Codes: B60N 2/002 Seats provided with an occupancy detection means mounted therein or thereon detection means responsive to presence or absence of children B60N2/266
• CPC Class Codes: B60R 21/0153 using field detection presence sensors
• CPC Class Codes: B60R 21/01532 using electric or capacitive field sensors
• CPC Class Codes: B60R 21/01556 Child-seat detection systems
• CPC Class Codes: G06V 20/59 inside of a vehicle, e.g. relating to seat occupancy, driver state or inner lighting conditions
• CPC Class Codes: G06V 40/103 Static body considered as a whole, e.g. static pedestrian or occupant recognition

Abstract

A seat occupant in a vehicle is classified into one of first and second categories by using a support vector machine (SVM). Sensor signals are obtained from seat sensors to form a sensor signal vector. By using the sensor signal vector, a value of a two-value decision function known as SVM is calculated. The seat occupant is decided to be of one of the first and second categories associated with the value. The two-value decision function can be generated through support vector learning by using a set of training vectors obtained from the seat sensors. An airbag system using an inventive seat occupant classifying apparatus is also disclosed.

Claims

1. A method of classifying a seat occupant or a state of the seat occupant into one of a first category and a second category in a vehicle having an array of sensors and a controller connected to the sensors for embodying the method, the method comprising the steps of:

(a) obtaining respective sensor signals from said sensors to form a sensor signal vector whose components are said sensor signals;

(b) calculating a value of a two-value decision function known as a support vector machine by using said sensor signal vector; and

(c) deciding said seat occupant to be of one of said first and second categories associated with said value.

2. A method as defined in claim 1, wherein said two-value decision function has been generated by using a set of first principal-component vectors obtained by a first principal component analysis of said set of training sensor signal vectors, said first principal-component vectors being smaller in dimension than said training sensor signal vectors, and wherein the method further comprises the step of performing a second principal component analysis of said sensor signal vector to obtain second principal-component vector and said calculating step using said second principal-component vector instead of said sensor signal vector.

3. A method as defined in claim 1, wherein said two-value decision function has been generated by using first dimension-reduced vectors whose dimension is lower than that of said sensor signal vector, said first dimension-reduced vectors being obtained by the steps of:

(d) applying an independent component analysis to each of training sensor signal vectors to obtain independent-component vectors;

(e) removing, from each independent-component vector, one of all components that provides a minimum classification error rate when said two-value decision function is generated by using said-one-component-removed versions of said independent-component vectors;

(f) repeating said step (e) till a dimension of said independent-component vectors becomes 1; and

(g) selecting one of component-removed versions that provides a minimum classification error rate as said first dimension-reduced vectors,

 

and wherein said method further including the steps of;

performing a second independent component analysis of said sensor signal vector to obtain second independent-component vector; and

form a second dimension-reduced vector that has the same dimension components as said first dimension-reduced vectors, and

said calculating step using said second dimension-reduced vector instead of said sensor signal vector.

4. A method as defined in claim 1, wherein said two-value decision function has been generated by using first independent-component vectors obtained by a first independent component analysis of each of training sensor signal vectors, and wherein said method further including the steps of:

performing a second independent component analysis of said sensor signal vector to obtain second independent-component vector; and

said calculating step using said second independent-component vector instead of said sensor signal vector.

5. A method as defined in claim 1, wherein said two-value decision function has been generated by using first dimension-reduced vectors whose dimension is lower than that of said sensor signal vector, said first dimension-reduced vectors being obtained by the steps of:

(d) applying a first principal component analysis to training sensor signal vectors to obtain first principal-component vectors, said first principal-component vectors being smaller in dimension than said training sensor signal vectors

(e) applying an independent component analysis to each of said first principal-component vectors to obtain independent-component vectors;

(f) removing, from each training sensor signal vector, one of all components that provides a minimum classification error rate when said two-value decision function is generated by using said-one-component-removed versions of said training sensor signal vectors;

(g) repeating said step (f) till a dimension of said independent-component vectors becomes 1; and

(h) selecting one of component-removed versions that provides a minimum classification error rate as said first dimension-reduced vectors,

 

and wherein said method further including the steps of;

performing a second principal component analysis of said sensor signal vector to obtain second principal-component vector;

performing a second independent component analysis of said second principal-component vector to obtain second independent-component vector; and

forming a second dimension-reduced vector that has the same dimension components as said first dimension-reduced vectors, and

said calculating step using said second dimension-reduced vector instead of said sensor signal vector.

6. A method as defined in claim 1, wherein said second category is classified into a first subcategory and a second subcategory and wherein the method further includes the steps of:

(b2) if said seat occupant is decided to be of said second category, calculating a second value of a second two-value decision function known as a support vector machine and associated with said second category by using said sensor signal vector; and

(c2) deciding said seat occupant to be of one of said first and second subcategories associated with said second value.

7. A method as defined in claim 6, wherein said first category is an adult class;

said second category is a child class;

said child class is divided into a sitting child class for a child seating on a vehicle seat and an infant seat class for an infant put in an infant seat.

8. A method as defined in claim 1, wherein said first and second categories relate to degrees of fatigue of said seat occupant.

9. A method as defined in claim 6, wherein said first category, said first subcategory and said second subcategory correspond to a slight, medium and heavy degrees of fatigue of said seat occupant.

10. A method as defined in claim 1, wherein said first and second categories relate to degrees of awakening of said seat occupant.

11. A method as defined in claim 8, further including the step of, if said seat occupant is decided to be of a higher degree, alerting at least said seat occupant.

12. A method as defined in claim 9, further including the step of, if said seat occupant is decided to be of said heavy degree, alerting at least said seat occupant.

13. A method as defined in claim 10, further including the step of, if said seat occupant is decided to be of a lower degree, alerting at least said seat occupant.

14. A method of generating a two-value decision function known as a support vector machine and used for classifying a seat occupant or a state of the seat occupant into one of a first category and a second category in a vehicle having an array of sensors and a controller connected to the sensors for effecting the classification, the method comprising the steps of:

(a) preparing a set of training input-output data samples, each training input-output data sample comprising a training sensor signal vector whose components are derived from said respective sensors and a class value that indicates said first or second category depending on said seat occupant at a time of obtaining said sensor signal vector; and

(b) using said set of training input-output data samples to find parameters for said two-value decision function through support vector learning.

15. A method as defined claim 14, further including the step of:

applying a principal component analysis to said set of training sensor signal vectors to obtain said set of principal-component vectors, said principal-component vectors being smaller in dimension than said training sensor signal vectors; and

said using step (b) using said set of principal-component vectors instead of said set of training input-output data samples.

16. A method as defined claim 14, further including the step of:

(c) applying an independent component analysis to each of training sensor signal vectors to obtain independent-component vectors;

(d) removing, from each independent-component vector, one of all components that provides a minimum classification error rate when said two-value decision function is generated by using said-one-component-removed versions of said independent-component vectors;

(e) repeating said step (d) till a dimension of said independent-component vectors becomes 1;

(f) selecting one of component-removed versions that provides a minimum classification error rate as dimension-reduced vectors,

(g) said using step (b) using said dimension-reduced vectors instead of said set of training input-output data samples.

17. A method as defined in claim 14, further including the step of:

(c) applying an independent component analysis to each of training sensor signal vectors to obtain independent-component vectors;

(d) said using step (b) using said independent-component vectors instead of said set of training input-output data samples.

18. A method as defined claim 14, further including the step of:

(d) removing, from each training sensor signal vector, one of all components that provides a minimum classification error rate when said two-value decision function is generated by using said-one-component-removed versions of said training sensor signal vectors;

(e) repeating said step (d) till a dimension of said training sensor signal vectors becomes 1;

(f) selecting one of component-removed versions that provides a minimum classification error rate as dimension-reduced vectors,

(g) said using step (b) using said dimension-reduced vectors instead of said set of training input-output data samples.

19. A method as defined claim 18, further including the step of removing sensors such that said array has only sensors that correspond to components of said dimension-reduced vectors.

20. An apparatus for classifying a seat occupant or a state of the seat occupant into one of a first category and a second category in a vehicle, the apparatus comprising:

an array of sensors each providing a sensor signal in response to an applied force; and

a controller connected to said sensors, said controller comprising;

(a) means for obtaining respective sensor signals from said sensors to form a sensor signal vector whose components are said sensor signals;

(b) means for calculating a value of a two-value decision function known as a support vector machine by using said sensor signal vector; and

(c) means for deciding said seat occupant to be of one of said first and second categories associated with said value.

21. An apparatus as defined in claim 20, further including:

means for performing a principal component analysis of said sensor signal vector to obtain principal-component vector, wherein

said calculating means includes means for using said principal-component vector instead of said sensor signal vector.

22. An apparatus as defined in claim 20, further including:

performing an independent component analysis of said sensor signal vector to obtain independent-component vector; and

form a dimension-reduced vector that has the same dimension components as said dimension-reduced vectors, wherein

said calculating means includes means for using said dimension-reduced vector instead of said sensor signal vector.

23. An apparatus as defined in claim 20, further including:

means for performing a independent component analysis of said sensor signal vector to obtain independent-component vector, wherein

said calculating means includes means for using said independent-component vector instead of said sensor signal vector.

24. An apparatus as defined in claim 20, further including:

means for performing a principal component analysis of said sensor signal vector to obtain principal-component vector;

means for performing an independent component analysis of said principal-component vector to obtain independent-component vector; and

means for forming a dimension-reduced vector that has the same dimension components as said dimension reduced vectors, wherein

said calculating means includes means for using said second dimension-reduced vector instead of said sensor signal vector.

25. An apparatus as defined in claim 20, wherein said second category is classified into a first subcategory and a second subcategory and wherein the apparatus further includes:

(b2) means, in the event said seat occupant is decided to be of said second category, for calculating a second value of a second two-value decision function known as a support vector machine and associated with said second category by using said sensor signal vector; and

(c2) means for deciding said seat occupant to be of one of said first and second subcategories associated with said second value.

26. An apparatus as defined in claim 25, wherein said first category is an adult class;

said second category is a child class;

said child class is divided into a sitting child class for a child seating on a vehicle seat and an infant seat class for an infant put in an infant seat.

27. An apparatus as defined in claim 20, wherein said first and second categories relate to degrees of fatigue of said seat occupant.

28. An apparatus as defined in claim 20, wherein said first and second categories relate to degrees of awakening of said seat occupant.

29. An apparatus as defined in claim 27, further including:

means, in the event said seat occupant is decided to be of a higher degree, for alerting at least said seat occupant.

30. An apparatus as defined in claim 28, further including:

means, in the event said seat occupant is decided to be of said lower degree, for alerting at least said seat occupant.

31. An apparatus for controlling airbag deployment in response to a state of a seat occupant in a vehicle, the apparatus comprising:

an array of force sensors each providing a sensor signal in response to an applied force; and

a first controller connected to said sensors for classifying said state of said seat occupant into one of a first category and a second category, said first controller comprising;

(a) means for obtaining respective sensor signals from said sensors to form a sensor signal vector whose components are said sensor signals;

(b) means for calculating a value of a two-value decision function known as a support vector machine by using said sensor signal vector;

(c) means for deciding said seat occupant to be of one of said first and second categories associated with said value;

means, in response to a detection of an imminent crash, for controlling airbag deployment on the basis of said one of said first and second categories.