Error detection in recognition data

• US 20170185868A1
• Filed: 12/24/2015
• Published: 06/29/2017
• Est. Priority Date: 12/24/2015
• Status: Active Grant

• Current Assignee: SAP SE
• Original Assignee: SAP SE
• Inventors: Zhang, Qiwei, Wang, Ke, Chen, Ling, Jin, Ye, Yu, Chao, Ni, Yinling, Liu, Jing, Li, Xin
• Granted Patent: US 10,163,038 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 16/583 using metadata automatically derived from the content
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06F 18/217 Validation; Performance evaluation; Active pattern learning techniques
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: G06V 10/774 Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06V 10/776 Validation; Performance evaluation
• CPC Class Codes: G06V 20/62 Text, e.g. of license plates, overlay texts or captions on TV images
• CPC Class Codes: G06V 20/625 License plates

Abstract

Described herein is a descriptive framework to facilitate error detection in recognition data. In accordance with one aspect of the framework, at least one erroneous record is detected in a first set of recognition data. The framework may determine a correction of a first recognized identifier in the erroneous record by searching a second set of recognition data for a matching record with a second recognized identifier substantially similar to the first recognized identifier. A report may then be generated to present the detected erroneous record and the determined correction.

Claims

1. A system for error detection, comprising:

a non-transitory memory device for storing computer-readable program code; and

a processor in communication with the memory device, the processor being operative with the computer-readable program code to perform operations comprising

preparing a training set based on at least a first set of recognition data associated with a first sensor, wherein the first set of recognition data comprises records that store recognized identifiers associated with vehicles,

training a confusion matrix based on the training set,

detecting at least one erroneous record in a second set of recognition data associated with the first sensor wherein the second set of recognition data comprises records that store recognized identifiers associated with the vehicles, and

determining correction of the erroneous record by comparing similarity with records associated with other sensors using the trained confusion matrix.

2. The system of claim 1 wherein the processor is operative with the computer-readable program code to prepare the training set by

identifying, based on at least one predetermined abnormal pattern, an erroneous record from the first set of recognition data that is likely to be falsely recognized;

and

pairing the identified erroneous record with a nearby matching record that provides a true value.

3. The system of claim 2 wherein the predetermined abnormal pattern comprises a one-time appearance of the erroneous record within a day.

4. The system of claim 2 wherein the processor is operative with the computer-readable program code to search for the nearby matching record based on a search criteria that ensures that the erroneous record and the matching record are captured within a predetermined time interval, the erroneous record and the matching record are captured by sensors that are within a predetermined distance, and the erroneous record and the matching record are substantially similar.

5. The system of claim 1 wherein the processor is operative with the computer-readable program code to prepare the training set further based on identification data associated with a second sensor.

6. The system of claim 5 wherein the identification data comprises radio frequency identification data associated with the vehicles.

7. The system of claim 5 further comprising multiple sensors in communication with the processor and the memory device that acquires the first and second sets of recognition data and the identification data.

8. The system of claim 5 wherein the processor is operative with the computer-readable program code to prepare the training set by

identifying, based on at least one predetermined abnormal pattern, an erroneous record from the first set of recognition data that is likely to be falsely recognized;

and

pairing the identified erroneous record with an identification data record that provides a true value.

9. The system of claim 8 wherein the processor is operative with the computer-readable program code to search for the identification data record based on a search criteria that ensures that the erroneous record and the identification data record are captured within a predetermined time interval, the erroneous record and the identification data record are captured by sensors that are within a predetermined distance, the erroneous record and the identification data record are associated with identifiers that are not exactly the same.

10. A computer-implemented method of error detection, comprising:

receiving a first set of recognition data associated with a sensor and a second set of recognition data associated with one or more other sensors;

detecting, based on a predetermined rule, at least one erroneous record in the first set of recognition data;

determining a correction of a first recognized identifier in the erroneous record by searching the second set of recognition data for a matching record with a second recognized identifier substantially similar to the first recognized identifier; and

generating a report that presents the detected erroneous record and the determined correction.

11. The method of claim 10 wherein receiving the first and second sets of recognition data comprises receiving records that store recognized registration identifiers of vehicles.

12. The method of claim 10 wherein detecting the at least one erroneous record comprises searching, in the first set of recognition data, for the erroneous record that occurred less than a predetermined threshold value in a predetermined time frame.

13. The method of claim 10 wherein determining the correction of the recognized identifier in the erroneous record comprises searching the second set of recognition data spatially and radially to find a matching record with a highest similarity score.

14. The method of claim 13 wherein searching the second set of recognition data comprises finding the matching record based on a search criteria that ensures that the erroneous record and the matching record are captured within a predetermined time interval, and the erroneous record and the matching record are captured by sensors that are within a predetermined distance.

15. The method of claim 10 wherein searching the second set of recognition data comprises determining a similarity based on a trained confusion matrix.

16. The method of claim 10 further comprising classifying a fault of the sensor based on an error distribution pattern associated with the erroneous record.

17. The method of claim 16 wherein the error distribution pattern represents a lane distribution of errors.

18. The method of claim 16 wherein the error distribution pattern represents a time distribution of errors.

19. The method of claim 16 wherein classifying the fault comprises classifying the fault into a type caused by illumination anomaly, special weather, obstruction of camera view or sensor malfunctioning.

20. A non-transitory computer-readable medium having stored thereon program code, the program code executable by a computer to perform steps comprising:

receiving a first set of recognition data associated with a sensor and a second set of recognition data associated with one or more other sensors;

detecting, based on a predetermined rule, at least one erroneous record in the first set of recognition data;

determining a correction of a first recognized identifier in the erroneous record by searching the second set of recognition data for a matching record with a second recognized identifier substantially similar to the first recognized identifier; and

generating a report that presents the detected erroneous record and the determined correction.