Method and apparatus for providing a low-power perception architecture

Patent metadata

Field	Value
Publication number	US-11144788-B2
Application number	US-201816209588-A
Country	US
Kind code	B2
Filing date	Dec 4, 2018
Priority date	Dec 4, 2018
Publication date	Oct 12, 2021
Grant date	Oct 12, 2021

Abstract

An approach is provided for providing a lower-power perception architecture. The approach involves, for example, determining that a device is equipped with a first perception system and a second perception system. The second perception system operates in a lower-power consumption mode than the first perception system to process image data for image recognition. The approach also involves determining a battery level of the device. The approach further involves switching from the first perception system to the second perception system based on determining that the battery level is below a threshold battery level.

First claim

What is claimed is:

1. A computer-implemented method for providing a low-power perception architecture comprising:
determining that a device is equipped with a first perception system and a second perception system, wherein the second perception system operates in a lower-power consumption mode than the first perception system to process image data for image recognition, and wherein the second perception system comprises a machine learning module trained to recognize objects in the image data;
determining a battery level of the device; and
switching from the first perception system to the second perception system based on determining that the battery level is below a threshold battery level.

2. The method of claim 1 , further comprising:
training the machine learning model of the second perception system using a cost function based on a power consumption of the machine learning model when used for the image recognition.

3. The method of claim 2 , wherein the cost function is further based on an accuracy of the machine learning model when performing the image recognition.

4. The method of claim 3 , wherein the machine learning model includes at least one input node, at least one hidden node, and at least one output node.

5. The method of claim 4 , further comprising:
evaluating a plurality of connections among the at least one input node, the at least one hidden node, the at least one output node, or a combination thereof against one or more levels of the power consumption, one or more levels of the accuracy, or a combination thereof.

6. The method of claim 5 , further comprising:
creating a plurality of perception architectures based on the plurality of connections corresponding to each of the one or more levels of the power consumption, the one or more levels of the accuracy, or a combination thereof.

7. The method of claim 6 , further comprising:
creating a stacked architecture comprising a plurality of layers,
wherein each of the plurality of layers is a respective one of the plurality of perception architectures.

8. The method of claim 7 , further comprising:
using a maximum pool layer of the stacked architecture to determine a consensus over respective outputs of said each of the plurality of layers.

9. The method of claim 8 , wherein the maximum pool layer has a maximum number of connections among the at least one input node, the at least one hidden node, the at least one output node, or a combination thereof.

10. The method of claim 1 , wherein the device is a vehicle or a component of the vehicle.

11. An apparatus for predicting sensor error, comprising:
at least one processor; and
at least one memory including computer program code for one or more programs,
the at least one memory and the computer program code configured to, with the at least one processor, cause the apparatus to perform at least the following,
determine that a device is equipped with a first perception system and a second perception system, wherein the second perception system operates in a lower-power consumption mode than the first perception system to process image data for image recognition, and wherein the second perception system comprises a machine learning module trained to recognize objects in the image data;
determine a battery level of the device; and
switch from the first perception system to the second perception system based on determining that the battery level is below a threshold battery level.

12. The apparatus of claim 11 , wherein the apparatus is further caused to:
train the machine learning model of the second perception system using a cost function based on a power consumption of the machine learning model when used for the image recognition.

13. The apparatus of claim 12 , wherein the cost function is further based on an accuracy of the machine learning model when performing the image recognition.

14. The apparatus of claim 13 , wherein the machine learning model includes at least one input node, at least one hidden node, and at least one output node.

15. The apparatus of claim 14 , wherein the apparatus is further caused to:
evaluate a plurality of connections among the at least one input node, the at least one hidden node, the at least one output node, or a combination thereof against one or more levels of the power consumption, one or more levels of the accuracy, or a combination thereof.

16. A non-transitory computer-readable storage medium for predicting sensor error, carrying one or more sequences of one or more instructions which, when executed by one or more processors, cause an apparatus to at least perform the following steps:
determining that a device is equipped with a first perception system and a second perception system, wherein the second perception system operates in a lower-power consumption mode than the first perception system to process image data for image recognition, and wherein the second perception system comprises a machine learning module trained to recognize objects in the image data;
determining a battery level of the device; and
switching from the first perception system to the second perception system based on determining that the battery level is below a threshold battery level.

17. The non-transitory computer-readable storage medium of claim 16 , wherein the apparatus is caused to further perform:
training the machine learning model of the second perception system using a cost function based on a power consumption of the machine learning model when used for the image recognition.

18. The non-transitory computer-readable storage medium of claim 17 , wherein the cost function is further based on an accuracy of the machine learning model when performing the image recognition.

19. The non-transitory computer-readable storage medium of claim 18 , wherein the machine learning model includes at least one input node, at least one hidden node, and at least one output node.

20. The non-transitory computer-readable storage medium of claim 19 , wherein the apparatus is caused to further perform:
evaluating a plurality of connections among the at least one input node, the at least one hidden node, the at least one output node, or a combination thereof against one or more levels of the power consumption, one or more levels of the accuracy, or a combination thereof.

Assignees

• Here Global Bv

Inventors

• Viswanathan Anirudh

Classifications

• G06V10/764 Primary

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

• G06V20/56

  exterior to a vehicle by using sensors mounted on the vehicle · CPC title

• G06V10/82

  using neural networks · CPC title

• G06F18/217

  Validation; Performance evaluation; Active pattern learning techniques · CPC title

• G06V10/94

  Hardware or software architectures specially adapted for image or video understanding · CPC title