Object detection system

Patent metadata

Field	Value
Publication number	US-12464963-B2
Application number	US-202318336681-A
Country	US
Kind code	B2
Filing date	Jun 16, 2023
Priority date	Jun 17, 2022
Publication date	Nov 11, 2025
Grant date	Nov 11, 2025

Abstract

An object detection system for an agricultural machine which utilises image data from one or more imaging sensors associated with the agricultural machine. The image data is analysed utilising first and second detection models to classify, for one or both models, an object within the environment of the agricultural machine. A classification metric for the object indicative of an overlap associated with the classification obtained for each of the models for the object is used in determining an identity for the object. One or more operable components associated with the agricultural machine may then be controlled in dependence on the determined identity.

First claim

What is claimed is:

1 . A control system for an object detection system for an agricultural machine, the control system comprising one or more controllers, and being configured to:
receive image data from one or more imaging sensors associated with the agricultural machine; analyze the image data utilizing first and second detection models to classify, for one or both models, an object within the environment of the agricultural machine; determine a classification metric for the object indicative of an overlap associated with the classification obtained for each of the models for the object; and determine an identity for the object in dependence on the classification metric and the classification obtained for the first and/or second models; wherein the control system is configured to:
control operation of one or more operable components associated with the agricultural machine in dependence on the determined identity; and
compare the classification metric with a threshold value.

2 . The control system of claim 1 , wherein the first and/or second detection models comprise machine-learned models trained on one or more training datasets with known objects with respective classifications.

3 . The control system as claimed in claim 2 , wherein the training datasets for first and second models are different.

4 . The control system as claimed in claim 3 , wherein the training dataset for the first model comprises an agricultural dataset; and the training dataset for the second model comprises a generic dataset.

5 . The control system of claim 1 , wherein classification by the first and/or second detection models comprises assignment of a class to the object from a plurality of classes associated with the model, and wherein the plurality of classes are be grouped by category, and optionally by subcategory.

6 . The control system as claimed in claim 5 , configured to determine the identity for the object in dependence on the class and/or category of the classification for the first and/or second model.

7 . The control system of claim 1 , wherein the classification output by the first and second models comprises a bounding box overlaid onto the image data at the location of the object as determined by the respective model; and the classification metric comprises a measure of an overlap of the bounding boxes determined for the first and second models.

8 . The control system as claimed in claim 1 , configured to determine the identity for the object in dependence on the classification for the first model only in dependence on the classification metric exceeding the threshold value.

9 . The control system as claimed in claim 1 , wherein the threshold value is variable, and is dependent on a context parameter.

10 . The control system as claimed claim 1 , wherein the threshold value is dependent on the classification assigned by the first model.

11 . The control system of claim 1 , wherein the one or more imaging sensors include a camera; and/or a LIDAR sensor.

12 . The control system of claim 1 , wherein the one or more operable components include a user interface; and the control system is operable to output, via the user interface, an indicator indicative of the determined identity to an operator of the agricultural machine.

13 . The control system of claim 1 , wherein the one or more operable components include a guidance system for the agricultural machine; and wherein the control system is configured to control operation of the guidance system for controlling operation thereof for controlling motion of the machine in dependence on the determined identity for the object.

14 . The control system of claim 1 , wherein the one or more operable components comprises a data server associated with the agricultural machine; and wherein the control system is operable to store on the data server information indicative of the determined identity for the object.

15 . An object detection system for an agricultural machine, comprising one or more imaging sensors; and wherein the system further comprises and/or is controllable by the control system of claim 1 .

16 . An agricultural machine comprising the control system of claim 1 .

17 . A method of object detection, comprising:
receiving image data from one or more imaging sensors associated with an agricultural machine; analyzing the image data utilizing first and second detection models to classify, for one or both models, an object within the environment of the agricultural machine; determining a classification metric for the object indicative of an overlap associated with the classification obtained for each of the models for the object; determining an identity for the object in dependence on the classification metric and the classification obtained for the first and/or second models; controlling operation of one or more operable components associated with the agricultural machine in dependence on the determined identity; and comparing the classification metric with a threshold value.

Assignees

• Agco Int Gmbh

Inventors

• Mujkic Esma

Classifications

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06V10/774

  Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting · 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

• G06V10/809

  of classification results, e.g. where the classifiers operate on the same input data · CPC title

• G06V20/64

  Three-dimensional [3D] objects · CPC title