Device for classifying data

Patent metadata

Field	Value
Publication number	US-11501162-B2
Application number	US-201716469246-A
Country	US
Kind code	B2
Filing date	Dec 8, 2017
Priority date	Dec 14, 2016
Publication date	Nov 15, 2022
Grant date	Nov 15, 2022

Abstract

A device is configured to classify data. Its operation involves providing (210) data samples including one or more of: image data, radar data, acoustic data, and/or lidar data to a processing unit. The data samples include at least one test sample, including positive samples and negative samples. Each positive sample has been determined to contain data relating to at least one object to be detected including at least one pedestrian, car, vehicle, truck or bicycle. Each negative sample has been determined not to contain data relating to the at least one object to be detected. These determinations regarding the positive samples and the negative samples are provided as input data, validated by a human operator, and/or provided by the device itself through a learning algorithm. A first plurality of groups is generated (220) by the processing unit implementing an artificial neural network, wherein at least some of the first plurality of groups are assigned a weighting factor. Each group of the first plurality of groups is populated (230) by the processing unit implementing the artificial neural network with a different at least one of the plurality of negative samples based on a different feature of sample data similarity for each group, which involves processing the negative samples to determine a number of different features of sample data similarity in order to populate different groups with negative samples that share or substantially share that or a similar feature, wherein at least one of the groups of the first plurality of groups contains at least two negative samples. It is determined (240) by the processing unit implementing the artificial neural network whether the at least one test sample contains data relating to the at least one object based on the plurality of the positive samples and the first plurality of groups. The artificial neural network implements the learning algorithm.

First claim

The invention claimed is:

1. A device ( 10 ) configured to classify data for determining driver warning information, comprising:
an input unit ( 20 ); and
a processing unit ( 30 );
wherein the input unit is configured to provide, to the processing unit, data samples comprising at least one of image data, radar data, acoustic data and/or lidar data obtained from a vehicle, wherein the data samples include at least one test sample including a plurality of positive samples and a plurality of negative samples, wherein a total number of the negative samples is significantly larger than a total number of the positive samples, wherein the plurality of negative samples includes a plurality of redundant samples with high similarity, wherein each one of the positive samples has been determined to contain data relating to at least one object to be detected including at least one pedestrian, car, vehicle, truck or bicycle, wherein each one of the negative samples has been determined not to contain data relating to the at least one object to be detected, and wherein the determination that the positive samples contain data relating to the at least one object to be detected and the determination that the negative samples do not contain data relating to the at least one object to be detected are provided as input data, validated by a human operator, and/or provided by the device through utilization of a learning algorithm;
wherein the processing unit ( 30 ) is configured to process the plurality of negative samples to determine a plurality of features of negative sample data similarity, and wherein the processing unit is configured to process the plurality of positive samples to determine a plurality of features of positive sample data similarity;
wherein the processing unit is configured to implement an artificial neural network to generate a first plurality of groups, wherein each one of the first plurality of groups is respectively assigned an importance weight or weighting factor, wherein during training a fine-grained balancing mechanism is applied with a respective group of the first plurality of groups being drawn based on the importance weight or weighting factor thereof, wherein when the respective group is drawn a sample from the respective group is drawn randomly, and wherein during the training a manual adjustment of the importance weights or weighting factors is performed to give higher importance to critical ones of the groups;
wherein the processing unit ( 30 ) is configured to determine a first plurality of representative samples for the first plurality of groups, wherein a different one of the representative samples is respectively associated with each respective group of the first plurality of groups;
wherein the processing unit is configured to implement the artificial neural network to populate each group of the first plurality of groups with a different at least one of the plurality of negative samples based on a different feature of sample data similarity for each one of the groups,
wherein the processing unit is configured to process the negative samples to determine a number of different features of sample data similarity in order to populate different ones of the groups with ones of the negative samples that share or substantially share that feature or a similar feature, wherein at least one of the groups of the first plurality of groups contains at least two of the negative samples, and wherein the processing unit is configured to populate each group of the first plurality of groups based on a different feature of negative sample data similarity for each one of the groups;
wherein the processing unit is configured to process a test sample of the at least one test sample to determine a test representative feature;
wherein the processing unit is configured to perform a selective addition that involves adding the test sample to the plurality of positive samples or adding the test sample to the plurality of negative samples, wherein the selective addition comprises a comparison of the test representative feature with the plurality of features of negative sample data similarity and the plurality of features of positive sample data similarity;
wherein the processing unit is configured to implement the artificial neural network to determine whether the at least one test sample contains the data relating to the at least one object based on the plurality of the positive samples and the first plurality of the representative samples; and
wherein the artificial neural network comprises the learning algorithm.

2. The device according to claim 1 , wherein the processing unit ( 30 ) is configured to generate a second plurality of groups, wherein the processing unit is configured to populate each group of the second plurality of groups with a different at least one of the plurality of positive samples, wherein at least one of the groups of the second plurality of groups contains at least two of the positive samples, and wherein the processing unit is configured to determine whether the at least one test sample contains the data relating to the at least one object based on the second plurality of groups and the first plurality of groups.

3. The device according to claim 2 , wherein at least some groups of the second plurality of groups are respectively assigned a respective weighting factor.

4. The device according to claim 2 , wherein the processing unit ( 30 ) is configured to populate each group of the second plurality of groups based on a different feature of sample data similarity for each group.

5. The device according to claim 2 , wherein the processing unit ( 30 ) is configured to determine a second plurality of representative samples for the second plurality of groups, wherein a different representative sample is associated with each group of the second plurality of groups, and wherein the processing unit is configured to determine whether the at least one test sample contains the data relating to the at least one object based on the second plurality of representative samples and the first plurality of representative samples.

6. The device according to claim 1 , wherein the processing unit ( 30 ) is configured to process the plurality of negative samples to determine a plurality of features of negative sample data similarity, wherein the processing unit is configured to populate each group of the first plurality of groups based on a different feature of negative sample data similarity for each group, wherein the processing unit is configured to process the plurality of positive samples to determine a plurality of features of positive sample data similarity, wherein the processing unit is configured to process the at least one test sample to determine at least one representative feature, and wherein the determination whether the at least one test sample contains the data relating to the at least one object comprises a comparison of the at least one representative feature with the plurality of features of negative sample data similarity and the plurality of features of positive sample data similarity.

7. The device according to claim 1 , wherein, when the processing unit ( 30 ) determines that the test sample is to be added to the plurality of negative samples, the processing unit is configured to add the test sample to a group of the first plurality of groups based on a similarity metric satisfying a threshold condition.

8. The device according to claim 1 , wherein, when the processing unit ( 30 ) determines that the test sample is to be added to the plurality of negative samples, the processing unit is configured to generate a new group of the first plurality of groups and add the test sample to the new group based on a similarity metric not satisfying a threshold

Assignees

• Conti Temic Microelectronic Gmbh

Inventors

• Karg Michelle
• Scharfenberger Christian
• Thiel Robert

Classifications

• G06V20/56

  exterior to a vehicle by using sensors mounted on the vehicle · CPC title

• G06F18/214 Primary

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

• G06F18/231

  Hierarchical techniques, i.e. dividing or merging pattern sets so as to obtain a dendrogram · CPC title

• G06F18/2178

  based on feedback of a supervisor · CPC title

• G06F18/2431

  Multiple classes · CPC title