Method and system for training and updating a classifier

Patent metadata

Field	Value
Publication number	US-11301721-B2
Application number	US-201716954099-A
Country	US
Kind code	B2
Filing date	Dec 15, 2017
Priority date	Dec 15, 2017
Publication date	Apr 12, 2022
Grant date	Apr 12, 2022

Abstract

Various embodiments of the teachings herein include a method for training and updating a backend-side classifier comprising: receiving, in a backend-device, from at least one vehicle, classification data along with a respective classification result generated by a vehicle-side classifier; and training the backend-side classifier using the classification data and, if available, a corrected respective classification result as annotation.

First claim

What is claimed is:

1. A method for training and updating a backend-side classifier, the method comprising:
receiving, in a backend-device, from at least one vehicle, classification data along with a respective classification result generated by a vehicle-side classifier; and
training the backend-side classifier using the classification data and, if available, a corrected respective classification result as annotation;
wherein training the backend-side classifier comprises:
comparing classification results for the same classification object received from a plurality of vehicles;
if the classification results differ, applying a consistency check on the classification results to determine an inconsistent or false classification result; and
training the backend-side classifier only using consistent annotations based on the consistency check.

2. The method of claim 1 , further comprising communicating a consistent annotation back to a vehicle in which an inconsistent or false classification result originates.

3. The method of claim 2 , further comprising initiating an update of the vehicle-side classifier of the vehicle in which the inconsistent or false classification result originates based on the consistent annotation.

4. The method of claim 1 , wherein applying a consistency check on the classification results comprises applying a voting scheme for classification results for the same classification object and setting the classification result that has been voted for by the voting scheme as a consistent annotation for the classification object.

5. The method of claim 1 , wherein applying a consistency check on the classification results comprises annotating the classification data by a manual annotator.

6. The method of claim 1 , wherein training the backend-side classifier comprises updating the backend-side classifier.

7. The method of claim 1 , wherein the backend-side classifier and the vehicle-side classifier is a multi-layer classifier having a plurality of layers and wherein the classification data is an output of a specific layer of the plurality of layers.

8. The method of claim 7 , wherein training the backend-side classifier comprises updating a number of layers corresponding to the specific layer of the plurality of layers.

9. A method for training and updating a backend-side classifier, the method comprising:
receiving, in a backend-device, from at least one vehicle, classification data along with a respective classification result generated by a vehicle-side classifier;
training the backend-side classifier using the classification data and, if available, a corrected respective classification result as annotation;
determining, in a vehicle, an inconsistent or false classification result;
generating, in the vehicle, an indication indicating that the classification is inconsistent or false;
receiving, in a backend-device, along with the classification data and the classification result the indication that the classification is inconsistent or false;
determining, by the backend-device, a correct classification result with respect to the classification data;
communicating, by the backend-device; the classification data and the correct classification result to the vehicle or another vehicle;
initiating, by the backend-device; a training of the vehicle-side classifier of the vehicle or the other vehicle based on the classification data and the determined correct classification result as annotation.

10. A method for training and updating a backend-side classifier, the method comprising:
receiving, in a backend-device, from at least one vehicle, classification data along with a respective classification result generated by a vehicle-side classifier;
training the backend-side classifier using the classification data and, if available, a corrected respective classification result as annotation;
determining, in a vehicle, an inconsistent or false classification result;
generating, in the vehicle, an indication indicating that the classification is inconsistent or false;
receiving, in a backend-device, along with the classification data and the classification result the indication that the classification is inconsistent false;
determining, by the backend-device, a correct classification result with respect to the classification data;
initiating, by the backend-device; a training of the backend-side classifier based on the classification data and the determined correct classification result as annotation.

11. A method for training and updating a backend-side classifier, the method comprising:
receiving, in a backend-device, from at least one vehicle, classification data along with a respective classification result generated by a vehicle-side classifier;
training the backend-side classifier using the classification data and, if available, a corrected respective classification result as annotation;
determining, in a vehicle, an inconsistent or false classification result for a classification object;
generating, in the vehicle, an indication indicating that the classification is inconsistent or false;
receiving, in a backend-device, along with the classification data and the classification result the indication that the classification is inconsistent false;
communicating, by the backend-device, the classification data to a plurality of other vehicles;
initiating, by the backend-device, a classification of the classification data by the vehicle-side classifier of each of the plurality of other vehicles;
receiving, by the backend-device, classification results from the plurality of other vehicles;
applying a voting scheme for the classification results and setting the classification result that the voting scheme has been voted for as a consistent annotation for the classification object;
communicating, by the backend-device, the consistent annotation to the vehicle or another vehicle; and
initiating, by the backend-device; a training of the vehicle-side classifier of the vehicle or another vehicle based on the classification data and the determined consistent annotation for the classification object.

12. A Backend-Device comprising:
a receiver configured to receive classification data along with a respective classification result generated by a vehicle-side classifier of at least one vehicle; and
one or more processors configured to implement a backend-side classifier and configured to train the backend-side classifier using the classification data and a possibly corrected respective classification result as annotation;
wherein the one or more processors are configured to compare classification results for the same classification object received from a plurality of vehicles; and
if the classification results differ, to apply a consistency check on the classification results to determine an inconsistent or false classification result; and to train the backend-side classifier only using consistent annotations based on the consistency check.

13. A system comprising:
a backend-device including a receiver configured to receive classification data along with a respective classification result generated by a vehicle-side classifier of at least one vehicle; and one or more processors configured to implement a backend-side classifier and configured to train the backend-side classifier using the classification data and a possibly corrected respective classification result as annotation; and
one or more vehicles each comprising a vehicle side classifier;
wherein the backend-device is further configured to communicate a consistent annotation back to a vehicle in which an inconsistent or false classification result originates and

Assignees

• Continental Automotive Gmbh

Inventors

• Hamer Henning
• Thiel Robert

Classifications

• G06V10/7796 Primary

  based on specific statistical tests · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06V20/582

  of traffic signs · CPC title

• G06F18/259

  Fusion by voting · CPC title

• G06V20/58

  Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads · CPC title