Neural network—instantiated lightweight calibration of RSS fingerprint dataset

• US 10,641,610 B1
• Filed: 06/03/2019
• Issued: 05/05/2020
• Est. Priority Date: 06/03/2019
• Status: Active Grant

• Current Assignee: Mapsted Corp.
• Original Assignee: Mapsted Corp.
• Inventors: Karon, Joshua, Ohab, Henry L., Huberman, Sean
• Primary Examiner(s): Badii, Behrang
• Assistant Examiner(s): Greene, Daniel L
• Application Number:
  US16/429,557
• Time in Patent Office: 337 Days
• CPC Class Codes: G01C 21/206 specially adapted for indoor navigation
• CPC Class Codes: G06N 3/084 Backpropagation, e.g. using gradient descent
• CPC Class Codes: H04W 4/024 Guidance services
• CPC Class Codes: H04W 4/027 using movement velocity, acceleration information
• CPC Class Codes: H04W 4/029 Location-based management or tracking services
• CPC Class Codes: H04W 4/33 for indoor environments, e.g. buildings
• CPC Class Codes: H04W 4/80 Services using short range communication, e.g. near-field communication [NFC], radio-frequency identification [RFID] or low energy communication

Abstract

A method and system of instantiating a lightweight re-calibration of a received signal strength (RSS) fingerprint dataset for mobile device indoor navigation. The method comprises, based on RSS parameters acquired from a plurality of mobile devices acquired at a set of positions within an indoor area, accumulating the RSS parameters in accordance with a trained neural network-based RSS fingerprint dataset in a fingerprint database of the indoor area; identifying respective positions of a subset of the set of positions having a variance that exceeds a threshold variance between observed RSS parameters and RSS parameters determined in accordance the trained neural network; and when contiguous positions of the subset are encompassed by a boundary representing a portion of the indoor area, instantiating a re-calibration of the RSS fingerprint dataset for mobile device navigation within the portion of the indoor area.

Claims

1. A method, executed in a processor of a server computing device, of instantiating a lightweight re-calibration of a received signal strength (RSS) fingerprint dataset for mobile device indoor navigation, the method comprising:

based on RSS parameters acquired from a plurality of mobile devices acquired at a set of positions within an indoor area, accumulating the RSS parameters in accordance with a trained neural network-based RSS fingerprint dataset in a fingerprint database of the indoor area;

identifying respective positions of a subset of the set of positions having a variance that exceeds a threshold variance between observed RSS parameters and RSS parameters determined in accordance the trained neural network; and

when contiguous positions of the subset are encompassed by a boundary representing a portion of the indoor area, automatically instantiating a re-calibration of the RSS fingerprint dataset for mobile device navigation within the portion of the indoor area, the portion having the variance exceeding the threshold variance.

2. The method of claim 1 wherein the portion of the indoor area comprises at least 10 percent of an area measurement of the indoor area.

3. The method of claim 1 wherein the threshold variance between the observed RSS parameters and the RSS parameters determined in accordance the trained neural network is at least 30%.

4. The method of claim 1 wherein the threshold variance persists for at least a predetermined duration of time.

5. The method of claim 1 wherein the trained neural network comprises a convolution neural network trained in accordance with:

computing, at an output layer of the convolution neural network implemented by the processor, an error matrix based on comparing an initial matrix of weights associated with the at least a first neural network layer representing a set of RSS input features to an RSS output feature in accordance with the RSS acquired parameters of the mobile device at the first location; and

recursively adjusting the initial weights matrix by backpropagation to diminish the error matrix until the RSS output feature matches the RSS acquired parameters;

wherein the backpropagation comprises a backward propagation of errors in accordance with the error matrix as computed at the output layer, the errors being distributed backwards throughout the weights of the at least one neural network layer.

6. The method of claim 1 wherein the indoor area is segmented into a plurality of functional areas, and the portion of the indoor area comprises at least 70 percent of an area measurement of at least one of the plurality of functional areas.

7. The method of claim 6 wherein the plurality of areas comprises a highly traversed area and a lesser traversed area.

8. The method of claim 7 wherein the threshold variance is a first threshold variance and further comprising a second threshold variance that is higher than the first threshold variance.

9. The method of claim 8 wherein the re-calibration is instantiated for the highly traversed area based on the first threshold variance, and instantiated for the lesser traversed area based on the second threshold variance.

10. The method of claim 8 wherein the more traveled and less traveled portions are determined based on accumulated historical data of mobile device traversals within the indoor area.

11. A server computing system for instantiating a lightweight re-calibration of a received signal strength (RSS) fingerprint dataset for mobile device indoor navigation, the server computing system comprising:

a processor; and

a memory including instructions executable in the processor to;

based on RSS parameters acquired from a plurality of mobile devices acquired at a set of positions within an indoor area, accumulate the RSS parameters in accordance with a trained neural network-based RSS fingerprint dataset in a fingerprint database of the indoor area;

identify respective positions of a subset of the set of positions having a variance that exceeds a threshold variance between observed RSS parameters and RSS parameters determined in accordance the trained neural network; and

when contiguous positions of the subset are encompassed by a boundary representing a portion of the indoor area, automatically instantiate a re-calibration of the RSS fingerprint dataset for mobile device navigation within the portion of the indoor area, the portion having the variance exceeding the threshold variance.

12. The system of claim 11 wherein the portion of the indoor area comprises at least 10 percent of an area measurement of the indoor area.

13. The system of claim 11 wherein the threshold variance persists for at least a predetermined duration of time.

14. The system of claim 11 wherein the trained neural network comprises a convolution neural network trained in accordance with instruction executable in the processor to:

compute, at an output layer of the convolution neural network implemented by the processor, an error matrix based on comparing an initial matrix of weights associated with the at least a first neural network layer representing a set of RSS input features to an RSS output feature in accordance with the RSS acquired parameters of the mobile device at the first location; and

recursively adjust the initial weights matrix by backpropagation to diminish the error matrix until the RSS output feature matches the RSS acquired parameters;

wherein the backpropagation comprises a backward propagation of errors in accordance with the error matrix as computed at the output layer, the errors being distributed backwards throughout the weights of the at least one neural network layer.

15. The system of claim 11 wherein the indoor area is segmented into a plurality of functional areas, and the portion of the indoor area comprises at least 70 percent of an area measurement of at least one of the plurality of functional areas.

16. The system of claim 15 wherein the plurality of areas comprises a highly traversed area and a lesser traversed area.

17. The system of claim 16 wherein the threshold variance is a first threshold variance and further comprising a second threshold variance that is higher than the first threshold variance.

18. The system of claim 17 wherein the re-calibration is instantiated for the highly traversed area based on the first threshold variance, and instantiated for the lesser traversed area based on the second threshold variance.

19. The system of claim 17 wherein the more traveled and less traveled portions are determined based on accumulated historical data of mobile device traversals within the indoor area.

20. The system of claim 17 wherein the threshold variance between the observed RSS parameters and the RSS parameters determined in accordance the trained neural network is at least 30%.