Method and device for classifying objects
A method for classifying objects which comprises a provision of measuring data from a sensor for a feature extraction unit as well as extraction of modality-independent features from the measuring data by means of the feature extraction unit, wherein the modality-independent features are independent of a sensor modality of the sensor, so that a conclusion to the sensor modality of the sensor is not possible from the modality-independent features.
1. A method for classifying objects, wherein the method comprises:
a provision of measuring data from a sensor for a feature extraction unit,
extraction of modality-independent features from the measuring data,
wherein the modality-independent features are independent of a sensor modality of the sensor, so that a conclusion to the sensor modality of the sensor is not possible from the modality-independent features,
wherein the extraction of modality-independent features from the measuring data comprises:
extracting features from the measuring data using the feature extractor unit comprising at least one feature extractor for each sensor modality,
transforming the features for each modality using a neural network of the feature extraction unit, and extracting by a feature transformation unit, the modality-independent features from the features of at least two feature extractors.
2. The method according to claim 1 ,
wherein there is at least a first sensor modality and a second sensor modality,
wherein the method is configured to extract the modality-independent features from the measuring data from a sensor of the first sensor modality in such a way that measuring data from a sensor of the second measuring modality can be reconstructed.
3. The method according to claim 1 ,
wherein the feature extractor is a neural network.
4. The method according to claim 1 ,
wherein the first sensor modality is lidar, and
wherein the feature extractor of the first sensor modality is a multi layer perceptron (MLP) network.
5. The method according to claim 1 ,
wherein the second sensor modality is an image, and
wherein the feature extractor of the second sensor modality is a convolutional neural network.
6. The method according to claim 1 ,
wherein the method comprises previous training of the feature extraction unit by means of unmonitored learning.
7. The method according to claim 1 ,
wherein the method comprises classification of the modality-independent features by means of a classification unit.
8. The method according to claim 7 ,
wherein the method comprises previous training of the classification unit by means of monitored learning.
9. The method according to claim 7 ,
wherein the method comprises
transferring of at least one feature vector from the feature extraction unit to the classification unit,
wherein the classification comprises comparison of the received feature vector to a respective previously determined average feature vector for each class, and
issuing a corresponding class label when falling below a previously specified deviation limit.
10. A device for carrying out a method for classifying objects which comprises:
provision of measuring data from a sensor for a feature extraction unit and extraction of modality-independent features from the measuring data,
wherein the modality-independent features are independent of a sensor modality of the sensor, so that a conclusion to the sensor modality of the sensor is not possible from the modality-independent features,
wherein the feature extraction unit comprises at least one feature extractor for each sensor modality,
wherein the feature extractor extracts features from the measuring data,
wherein the feature extraction unit comprises a feature transformation unit,
wherein the feature transformation unit comprises a neural network for each sensor modality, and wherein the feature transformation unit extracts the modality-independent features from the features of at least two feature extractors.
11. A non-transitory computer-readable storage medium on which a program is stored, which when executed by a computer carries out a method for classifying objects, said method comprises:
a provision of measuring data from a sensor for a feature extraction unit, and
extraction of modality-independent features from the measuring data,
wherein the modality-independent features are independent of a sensor modality of the sensor, so that a conclusion to the sensor modality of the sensor is not possible from the modality-independent features,
wherein the extraction of modality-independent features from the measuring data comprises:
extracting features from the measuring data using the feature extractor unit comprising at least one feature extractor for each sensor modality,
transforming the features for each sensor modality using a neural network of the feature extraction unit, and
extracting by a feature transformation unit, the modality-independent features from the features of at least two feature extractors.