Method and apparatus for distributed edge learning

• US 10,489,693 B2
• Filed: 08/03/2017
• Issued: 11/26/2019
• Est. Priority Date: 08/03/2017
• Status: Active Grant

• Current Assignee: Motorola Solutions, Inc.
• Original Assignee: Motorola Solutions, Inc.
• Inventors: Sabripour, Shervin
• Primary Examiner(s): Akhavannik, Hadi
• Application Number:
  US15/668,624
• Publication Number:
  US 20190042896A1
• Time in Patent Office: 845 Days
• Field of Search: None
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06V 10/82 using neural networks
• CPC Class Codes: G06V 20/56 exterior to a vehicle by using sensors mounted on the vehicle

Abstract

An object learning system, method, and device. The object learning device includes an electronic processor configured to provide an identifier based on target to at least one auxiliary object learning device and initiate an edge learning process on the target to create first preprocessed object recognition data. The electronic processor is further configured to receive second preprocessed object recognition data corresponding to the target from the at least one auxiliary object learning device and create, based on the first and the second preprocessed object recognition data, a classifier of the target.

Claims

1. An object learning device configured to operate at an edge of a network, the object learning device comprising:

a camera;

an electronic processor configured to

detect a target within a first field of view of the camera;

provide an identifier, based on the target, to at least one auxiliary object learning device,

initiate an object learning process on the target to create first preprocessed object recognition data,

receive, from the at least one auxiliary object learning device, second preprocessed object recognition data corresponding to the target within a second field of view of a second camera of the at least one auxiliary object learning device,

create, based on the first and the second preprocessed object recognition data, a classifier of the target, and

identify an object based on the classifier.

2. The object learning device of claim 1, wherein the electronic processor is further configured to assign a role to the at least one auxiliary object learning device based on at least one selected from the group consisting of a processing ability of the auxiliary object learning device, a quality of view of the target, and an amount of additional object recognition data that is able to be collected.

3. The object learning device of claim 1, wherein the object learning device is assigned a master role and the electronic processor is further configured to transfer the master role to the at least one auxiliary object learning device when the object learning device is unable to detect the target or no additional object recognition data is collected by the object learning device.

4. The object learning device of claim 3, wherein the electronic processor is further configured to select the at least one auxiliary object learning device from a plurality of auxiliary object learning devices to transfer the master role to.

5. The object learning device of claim 1, wherein the at least one auxiliary object learning device is a remote server.

6. A method of object learning on an edge of a distributed network, the method comprising:

detecting, via a camera, a target within a first field of view of the camera;

providing, via a master node, an identifier, based on the target, to at least one auxiliary node;

initiating, via an electronic processor of the master node, an object learning process on the target, creating a first preprocessed object recognition data;

receiving, at the master node, a second preprocessed object recognition data corresponding to the target within a second field of view of a second camera of the at least one auxiliary node;

creating, via the electronic processor, based on the first and the second object recognition data, a classifier of the target; and

identifying an object based on the classifier.

7. The method of claim 6, the method further comprising assigning a role to the at least one auxiliary node based on at least one selected from the group consisting of a processing ability of the auxiliary node, a quality of view of the target, and an amount of additional object recognition data that is able to be collected.

8. The method of claim 7, wherein the master node is assigned a master role and the method further comprising transferring the master role to the at least one auxiliary node when the master node is unable to detect the target or no additional object recognition data is collected by the master node.

9. The method of claim 8, the method further comprising determining, from a plurality of auxiliary nodes, an optimal auxiliary node to transfer the master role to.

10. The method of claim 6, wherein the at least one auxiliary node is a remote server.

11. An object learning system comprising:

an auxiliary object learning device including a first camera and a first electronic processor, the first electronic processor configured to

detect a target within a first field of view of the first camera;

initiate a first object learning process on the target to create first preprocessed object recognition data; and

an object learning device including a second camera and a second electronic processor configured to

detect the target within a second field of view of the second camera;

provide an identifier, based on the target, to the auxiliary object learning device,

initiate a second object learning process on the target to create second preprocessed object recognition data,

receive first preprocessed object recognition data corresponding to the target from the auxiliary object learning device,

create, based on the first and the second preprocessed object recognition data, a classifier of the target, and

identify an object based on the classifier.

12. The object learning system of claim 11, wherein the second electronic processor is further configured to assign a role to the auxiliary object learning device based on a processing ability of the auxiliary object learning device, a quality of view of the target, and an amount of additional object recognition data that is able to be collected.

13. The object learning system of claim 12, wherein the object learning device is assigned a master role and the second electronic processor is further configured to transfer the master role to the auxiliary object learning device when the object learning device is unable to detect the target or no additional object recognition data is collected by the object learning device.

14. The object learning system of claim 13, wherein the second electronic processor is further configured to determine from a plurality of auxiliary object learning devices, including the auxiliary object learning device, one of the plurality of auxiliary object learning devices to transfer the master role to.

15. The object learning system of claim 11, wherein the auxiliary object learning device is a remote server.