Method, device, computer program, and machine-readable storage medium for the detection of an object

Patent metadata

Field	Value
Publication number	US-11455791-B2
Application number	US-202016925454-A
Country	US
Kind code	B2
Filing date	Jul 10, 2020
Priority date	Jul 22, 2019
Publication date	Sep 27, 2022
Grant date	Sep 27, 2022

Abstract

A method for the detection of an object in an environment of a vehicle as a function of sensor signals of a sensor for acquiring the environment of the vehicle. The method includes: processing the sensor signals using a region proposal network to obtain at least one object hypothesis per anchor, the object hypothesis including an object probability and a bounding box; selecting the best object hypothesis on the basis of a quality model, the quality model being a function of the anchor and the bounding box of the object hypothesis; identifying redundant object hypotheses relative to the selected object hypothesis, the redundant object hypotheses being identified as a function of the anchors of the redundant object hypotheses, using a target function assigned to the region proposal network; and fusing the selected object hypothesis with the identified redundant object hypotheses for the object detection.

First claim

What is claimed is:

1. A method for detection of objects in an environment of a vehicle as a function of sensor signals of a sensor, the sensor signals representing the environment of the vehicle, the method comprising:
processing the sensor signals using a region proposal network that defines a plurality of anchors for a region of the environment, wherein the processing includes, for each of at least one of the anchors:
obtaining a plurality of object hypotheses for the respective anchor, each respective one of the plurality of object hypotheses specifying a bounding box of a respective object and a respective probability of presence of the respective object, whose bounding box is specified by the respective object hypothesis, is located at the respective anchor;
selecting a best one of the plurality of object hypotheses based on relationships of the anchor and the respective bounding boxes of the object hypotheses;
based on the selection of the best one of the plurality of object hypotheses, using a relationship of the selected best one of the plurality of object hypotheses to each of all others of the plurality of object hypotheses to identify a subset of the others of the plurality of object hypotheses that are relevant to the selected best one of the plurality of object hypotheses; and
based on the identification, fusing the selected best one of the plurality of object hypotheses with the identified object hypotheses.

2. The method as recited in claim 1 , wherein the identification is determined based on respective distances of the bounding box defined by the selected best one of the plurality of object hypotheses from the respective bounding boxes of the others of the plurality of object hypotheses.

3. The method as recited in claim 1 , wherein the fusion is performed based on respective quality measures of the selected best one of the plurality of object hypotheses and of the identified subset of the others of the plurality of object hypotheses, each of the quality measures being a function of a respective relative position of the anchor relative to the respective bounding box of the respective object hypothesis.

4. The method as recited in claim 3 , wherein the fusion is performed based on respective quality measures of the selected best one of the plurality of object hypotheses and of the identified subset of the others of the plurality of object hypotheses, and the quality measures are a function of the region proposal network.

5. The method as recited in claim 3 , wherein the fusion is performed based on respective quality measures of the selected best one of the plurality of object hypotheses and of the identified subset of the others of the plurality of object hypotheses, and the quality measures are a function of a geometry between the sensor and the object of the object hypotheses.

6. The method as recited in claim 1 , wherein the fusion produces a new version of the selected object hypothesis, and the method further comprises re-performing the identification using the new version of the selected object hypothesis.

7. The method as recited in claim 1 , wherein the fusion is performed based on respective quality measures of the selected best one of the plurality of object hypotheses and of the identified subset of the others of the plurality of object hypotheses, each of the quality measures being a function of a respective covariance matrix of parameters of the respective bounding box of the respective object hypothesis.

8. The method as recited in claim 1 , further comprising determining, based on a result of the fusion, that the selected best one of the plurality of object hypotheses is a false positive indication of the presence of the respective object.

9. A device configured for detection of objects in an environment of a vehicle as a function of sensor signals of a sensor, the sensor signals representing the environment of the vehicle, the device configured to:
process the sensor signals using a region proposal network that defines a plurality of anchors for a region of the environment, wherein the processing of the sensor signals includes, for each of at least one of the anchors:
obtaining a plurality of object hypotheses for the respective anchor, each respective one of the plurality of object hypotheses specifying a bounding box of a respective object and a respective probability of presence of the respective object, whose bounding box is specified by the respective object hypothesis, is located at the respective anchor;
selecting a best one of the plurality of object hypotheses based on relationships of the anchor and the respective bounding boxes of the object hypotheses;
based on the selection of the best one of the plurality of object hypotheses, using a relationship of the selected best one of the plurality of object hypotheses to each of all others of the plurality of object hypotheses to identify a subset of the others of the plurality of object hypotheses that are relevant to the selected best one of the plurality of object hypotheses; and
based on the identification, fusing the selected best one of the plurality of object hypotheses with the identified object hypotheses.

10. A non-transitory machine-readable storage medium on which is stored a computer program for detection of objects in an environment of a vehicle as a function of sensor signals of a sensor, the sensor signals representing the environment of the vehicle, the computer program, when executed by a computer, causing the computer to perform the following steps:
processing the sensor signals using a region proposal network to that defines a plurality of anchors for a region of the environment, wherein the processing includes, for each of at least one of the anchors:
obtaining a plurality of object hypotheses for the respective anchor, each respective one of the plurality of object hypotheses specifying a bounding box of a respective object and a respective probability of presence of the respective object, whose bounding box is specified by the respective object hypothesis, is located at the respective anchor;
selecting a best one of the plurality of object hypotheses based on relationships of the anchor and the respective bounding boxes of the object hypotheses;
based on the selection of the best one of the plurality of object hypotheses, using a relationship of the selected best one of the plurality of object hypotheses to each of all others of the plurality of object hypotheses to identify a subset of the others of the plurality of object hypotheses that are relevant to the selected best one of the plurality of object hypotheses; and
based on the identification, fusing the selected best one of the plurality of object hypotheses with the identified object hypotheses.

Assignees

• Bosch Gmbh Robert

Inventors

• Faion Florian
• Condurache Alexandru Paul
• Glaeser Claudius
• Drews Florian
• Ebert Jasmin
• Rosenbaum Lars
• Stal Rainer
• Muenzner Sebastian
• Gumpp Thomas

Classifications

• G06V10/255

  Detecting or recognising potential candidate objects based on visual cues, e.g. shapes · CPC title

• G06V10/82

  using neural networks · CPC title

• G06V10/454

  Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN] · CPC title

• G06N3/08 Primary

  Learning methods · CPC title

• G06F18/25 Primary

  Fusion techniques · CPC title