Method and device for determining a classification for an object based on a radar spectrum and radar reflections

Patent metadata

Field	Value
Publication number	US-12535558-B2
Application number	US-202318522469-A
Country	US
Kind code	B2
Filing date	Nov 29, 2023
Priority date	Dec 5, 2022
Publication date	Jan 27, 2026
Grant date	Jan 27, 2026

Abstract

A device and method for determining a classification of an object. A radar spectrum which includes radar reflections is determined as a function of sensor data from a radar sensor. Embeddings of sections from the radar spectrum which include at least one radar reflection are determined and each is assigned to a node of a first graph. Edges are determined as a function of pairwise distances between the embeddings which are assigned to the nodes. For each edge, a feature for the edge is determined as a function of the embeddings of two nodes which are connected to one another by the edge. For each node, a feature for the node is determined as a function of the features for the edges which connect the node to another node. A characteristic quantity is provided for each radar reflection as a function of the sensor data.

First claim

What is claimed is:

1 . A method for determining a classification of an object, comprising the following steps:
acquiring sensor data using a radar sensor; determining a radar spectrum which includes radar reflections as a function of the sensor data; determining embeddings of sections from the radar spectrum which include at least one radar reflection, which are each assigned to a node of a first graph; determining edges of the first graph as a function of pairwise distances between the embeddings which are assigned to the nodes of the first graph; for each edge of the first graph, determining a feature for the edge of the first graph as a function of the embeddings of two nodes of the first graph which are connected to one another by the edge of the first graph; for each node of the first graph, determining a feature for the node of the first graph as a function of the features for the edges of the first graph that connect the node of the first graph in the first graph to another node of the first graph; providing characteristic quantities, including spatial coordinates or at least one measured property, for each radar reflection as a function of the sensor data; for each radar reflection, determining an embedding which is assigned to a node of a second graph and which includes an embedding of the characteristic quantities and the characteristic quantities; determining edges of the second graph as a function of pairwise distances between the embeddings which are assigned to the nodes of the second graph; for each edge of the second graph, determining a feature for the edge of the second graph as a function of the embeddings of two nodes of the second graph which are connected to one another by the edge of the second graph; for each node of the second graph, determining a feature for the node of the second graph as a function of the features for the edges of the second graph which connect the node of the second graph in the second graph to another node of the second graph; and determining the classification of the object as a function of a quantity which includes a first part which is determined as a function of the features for the nodes of the first graph, and which includes a second part which is determined as a function of the features for the nodes of the second graph.

2 . The method as recited in claim 1 , wherein the embeddings of the sections from the radar spectrum include embedded features, the embedded features being extracted from spectral features from the sections using an encoder.

3 . The method as recited in claim 1 , characterized in that the first graph and/or the second graph is formed in a feature space extended by the features.

4 . The method as recited in claim 1 , wherein the quantity includes a third part: (i) which includes the embeddings of the nodes of the first graph, and/or (ii) which includes the embeddings of the nodes of the second graph.

5 . The method as recited in claim 1 , wherein: (i) the pairwise distance of embeddings which are each assigned to one of the nodes of the first graph is determined as a function of a difference between the embeddings, and/or (ii) the pairwise distance of embeddings which are each assigned to one of the nodes of the second graph is determined as a function of a difference between the embeddings.

6 . The method as recited in claim 1 , wherein, for each node of the first graph, the feature for the node of the first graph is determined as a function of a distance from the radar sensor and a Doppler velocity which is assigned to a radar reflection which includes the section of the radar spectrum assigned to the node of the first graph.

7 . The method as recited in claim 1 , wherein the first graph includes edges which each connect one of the nodes of the first graph to a specified number of its closest nodes in the first graph, or the first graph includes edges which each connect one of the nodes of the first graph to a specified number of its most distant nodes of the first graph, or the first graph includes edges which each connect one of the nodes of the first graph to a specified number of randomly selected other nodes of the first graph, or the first graph includes edges by which each node in the first graph is connected to every other node of the first graph by an edge.

8 . The method as recited in claim 1 , wherein the second graph includes edges which connect each of the nodes to a specified number of its closest nodes in the second graph, or the second graph includes edges which connect each of the nodes of the second graph to a specified number of its most distant nodes of the second graph, or the second graph includes edges which each connect one of the nodes of the second graph to a specified number of randomly selected other nodes of the second graph, or the second graph includes edges by which each node in the second graph is connected to every other node of the second graph by an edge.

9 . A device for determining a classification of an object, the device comprising:
at least one memory, and at least one processor, the at least one processor being configured to execute machine-readable instructions, and the at least one memory is configured to store the instruction; wherein the instructions, when executed by the at least one processor, causes the at least one processor to perform:
acquiring sensor data using a radar sensor,
determining a radar spectrum which includes radar reflections as a function of the sensor data,
determining embeddings of sections from the radar spectrum which include at least one radar reflection, which are each assigned to a node of a first graph,
determining edges of the first graph as a function of pairwise distances between the embeddings which are assigned to the nodes of the first graph,
for each edge of the first graph, determining a feature for the edge of the first graph as a function of the embeddings of two nodes of the first graph which are connected to one another by the edge of the first graph,
for each node of the first graph, determining a feature for the node of the first graph as a function of the features for the edges of the first graph that connect the node of the first graph in the first graph to another node of the first graph,
providing characteristic quantities, including spatial coordinates or at least one measured property, for each radar reflection as a function of the sensor data,
for each radar reflection, determining an embedding which is assigned to a node of a second graph and which includes an embedding of the characteristic quantities and the characteristic quantities,
determining edges of the second graph as a function of pairwise distances between the embeddings which are assigned to the nodes of the second graph,
for each edge of the second graph, determining a feature for the edge of the second graph as a function of the embeddings of two nodes of the second graph which are connected to one another by the edge of the second graph,
for each node of the second graph, determining a feature for the node of the second graph as a function of the features for the edges of the second graph which connect the node of the second graph in the second graph to another node of the second graph, and
determining the classification of the object as a function of a quantity which includes a first part which is determined as a function of the features for the nodes of the first graph, and which includes a second part which is determined as a function of the features for the nodes of the second graph.

10 . The device as recited in claim 9 , further comprising the radar sensor configured to acquire the sensor data.

11 . A non-transitory m

Assignees

• Bosch Gmbh Robert

Inventors

• Hakobyan Gor
• Acosta Aponte Axel Javier
• Saini Loveneet

Classifications

• G01S13/42

  Simultaneous measurement of distance and other co-ordinates (indirect measurement G01S13/46) · CPC title

• G01S7/411 Primary

  Identification of targets based on measurements of radar reflectivity (G01S7/415 takes precedence) · CPC title

• G01S13/726

  Multiple target tracking · CPC title

• G01S13/58

  Velocity or trajectory determination systems; Sense-of-movement determination systems · CPC title

• G01S13/88

  Radar or analogous systems specially adapted for specific applications (electromagnetic prospecting or detecting of objects, e.g. near-field detection, G01V3/00) · CPC title