Method for evaluation and stabilization over time of classification results

• US 7,796,820 B2
• Filed: 09/03/2004
• Issued: 09/14/2010
• Est. Priority Date: 10/02/2003
• Status: Active Grant

• Current Assignee: Robert Bosch GmbH
• Original Assignee: Robert Bosch GmbH
• Inventors: Simon, Stephan, Voelz, Henning, Lindner, Sebastian
• Primary Examiner(s): Mariam; Daniel G
• Assistant Examiner(s): Woldemariam; Aklilu k
• Application Number:
  US10/573,221
• Publication Number:
  US 20070036428A1
• Time in Patent Office: 2,202 Days
• Field of Search: 382/159, 382/224, 382/118, 382/143, 367/135, 701/301, 702/185, 706/20, 709/200, 324/457, 707/4, 715/234
• US Class Current: 382/224
• CPC Class Codes: G06V 20/52 Surveillance or monitoring of activities, e.g. for recognising suspicious objects recognising microscopic objects G06V20/69
• CPC Class Codes: G08G 1/017 identifying vehicles G08G1/015, G08G1/054 take precedence

Abstract

Classification methods are described that proceed in computer-assisted fashion, and in particular a method for evaluation and stabilization over time of classification results is described in which objects to be classified are sensed using sensors over a period of time, and are repeatedly classified with the inclusion of specific quality parameters for each object class. To ensure better classification reliability, the following steps may be carried out: a) increasing the value of the confidence parameter if a subsequent classification confirms the result of a previous classification; b) decreasing the value of the confidence parameter if a subsequent classification does not confirm the result of a previous classification; and c) generating a final classification result including the confidence parameters that have been increased or decreased in value.

Claims

1. A method for evaluation and stabilization over time of classification results from a classification method which proceeds in computer-assisted fashion, the method comprising:

sensing, by sensors, objects to be classified over a period of time;

repeatedly classifying the objects, by a computer processor, using specific quality parameters for each object class, wherein each classification utilizes a separate instance of sensor data obtained over the period of time by the sensors;

increasing, by the processor, a value of a confidence parameter of an object class, the confidence parameter being calculated from the quality parameters and the increasing being conditional upon a subsequent one of the classifications confirming a result of a previous one of the classifications;

decreasing, by the processor, the value of the confidence parameter of an object class, the decreasing being conditional upon a subsequent one of the classifications failing to confirm the result of a previous one of the classifications; and

generating, by the processor, a final classification result including the confidence parameters that have been increased or decreased in value.

2. The method as recited in claim 1, wherein the increasing of the value is performed as a function of an absolute quality of the confidence parameter.

3. The method as recited in claim 1, wherein the decreasing in the value is performed as a function of an absolute quality of the confidence parameter.

4. The method as recited in claim 1, wherein an absolute quality of respective individual results of the classification method is included in at least one of the increase in the value of the respective confidence parameters, and the decrease in the value of the respective confidence parameter, in weighted fashion with reference to individual object classes.

5. The method as recited in claim 1, further comprising:

limiting a permissible value range for the confidence parameters increased or decreased in value.

6. The method as recited in claim 1, further comprising:

evaluative analyzing the calculated confidence parameter to determine a final, detailed classification result.

7. The method as recited in claim 6, further comprising:

evaluating an alternation of the classification results between specific object classes as a classification into a higher-order class than those object classes.

8. The method as recited in claim 6, further comprising:

evaluating an alternation of the classification results between dissimilar object classes as a rejection of a classification of the object.

9. The method as recited in claim 1, further comprising:

evaluating classification results of the classification method for objects in surroundings of a vehicle.

12. The method as recited in claim 1, wherein the confidence parameters are sorted into a list according to value.

14. The method as recited in claim 12, wherein conditional upon the confidence parameters calculated during a classification indicating that an object does not belong to any class, the confidence parameters of all classes are decreased in value.

15. The method as recited in claim 1, wherein the processor rejects the final classification result as invalid conditional upon the confidence parameter with the highest value failing to exceed a predetermined minimum threshold value.

10. A computer-assisted vehicle information system, comprising:

connection interfaces to vehicle sensor devices for sensing objects in surroundings of a vehicle; and

a control circuit configured to analyze and classify the sensed objects the control circuit configured to perform the following;

sensing objects to be classified using sensors over a period of time;

repeatedly classifying the objects using specific quality parameters for each object class, wherein each classification utilizes a separate instance of sensor data obtained over the period of time by the sensors;

increasing a value of a confidence parameter of an object class, the confidence parameter being calculated from the quality parameters and the increasing being conditional upon a subsequent one of the classifications confirming a result of a previous one of the classifications;

decreasing the value of the confidence parameter of an class, the decreasing being conditional upon a subsequent one of the classifications failing to confirm the result of a previous one of the classifications; and

generating a final classification result including the confidence parameters that have been increased or decreased in value.

11. The vehicle information system as recited in claim 10, further comprising:

interfaces connected to actuator devices on the vehicle.

13. The method as recited in claim 10, wherein the increasing includes increasing the value of the confidence parameters by varying amounts based on a list rank of each confidence parameter.

16. The method as recited in claim 11, wherein the processor rejects the final classification result as invalid conditional upon a difference between the values of a highest confidence parameter and a second-highest confidence parameter failing to exceed a predetermined minimum threshold value.