Magnetic fingerprint neural network training for mobile device indoor navigation
A method and system of magnetic fingerprint based neural network training for mobile device indoor navigation and positioning. The method, executed in a processor of a server computing device, comprises determining, in the processor, at a plurality of locations, a set of magnetic input parameters in accordance with a magnetic infrastructure profile of at least a portion of an indoor area, the processor implementing an input layer of a neural network, the set of magnetic input parameters providing a magnetic feature input to the input layer of the neural network; receiving, from a mobile device positioned at the first location, a set of measured magnetic parameters at respective ones of the plurality of locations; computing, at an output layer of the 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 the magnetic feature input to a magnetic feature output in accordance with the magnetic measured parameters of the mobile device; and recursively adjusting the initial weights matrix by backpropogation to diminish the error matrix until the generated magnetic feature output matches the magnetic measured parameters.
1. A method, executed in a processor of a server computing device, for mobile device indoor navigation and positioning, the method comprising:
training a neural network comprising,
determining, in the processor, at a plurality of locations, a set of magnetic input parameters in accordance with a magnetic infrastructure profile of at least a portion of an indoor area, the processor implementing an input layer of a neural network, the set of magnetic input parameters providing a magnetic feature input to the input layer of the neural network, wherein the magnetic infrastructure profile is based on potential magnetic field interference sources in the indoor area;
receiving, from a mobile device positioned at a first location, a set of measured magnetic parameters at respective ones of the plurality of locations;
computing, at an output layer of the 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 the magnetic feature input to a magnetic feature output in accordance with the magnetic measured parameters of the mobile device; and
recursively adjusting the initial weights matrix by backpropogation to diminish the error matrix until the generated magnetic feature output matches the magnetic measured parameters;
generating a magnetic fingerprint map in the indoor area based on trained neural network; and
localizing the mobile device in the indoor area based on the magnetic fingerprint map.
2. The method of claim 1 wherein the set of magnetic input parameters and the initial weights matrix are determined in the processor of the server computing device in accordance with execution of the magnetic infrastructure profile.
3. The method of claim 1 wherein the set of magnetic measured parameters comprises a magnetic field strength.
4. The method of claim 1 wherein the set of magnetic measured parameters comprises a magnetic dip angle.
5. The method of claim 1 wherein the magnetic infrastructure profile comprises at least one of a steel structural element and a ferro-magnetic structural element.
6. The method of claim 1 wherein a magnetic field profile is determined in accordance with a postulated mathematical magnetic model to determine default initial values of the set of magnetic input parameters.
7. The method of claim 1 wherein the neural network is one of a recurrent neural network and a convolution neural network.
8. The method of claim 7 wherein the neural network comprises the convolution neural network, wherein the at least a first neural network layer corresponds to the set of magnetic input parameters for a magnetic field component.
9. The method of claim 1 further comprising recursively adjusting the initial weights matrix as the error matrix is diminished until the generated magnetic feature output matches the magnetic measured parameters within a threshold percentage value of the magnetic measured parameters.
10. The method of claim 1 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.
11. A server computing system for mobile device indoor navigation and positioning, the server computing system comprising:
a processor; and
a memory including instructions executable in the processor to:
train a neural network comprising,
determine, in the processor, at a plurality of locations, a set of magnetic input parameters in accordance with a magnetic infrastructure profile of at least a portion of an indoor area, the processor implementing an input layer of a neural network, the set of magnetic input parameters providing a magnetic feature input to the input layer of the neural network, wherein the magnetic infrastructure profile is based on potential magnetic field interference sources in the indoor area;
receive, from a mobile device positioned at the first location, a set of measured magnetic parameters at respective ones of the plurality of locations;
compute, at an output layer of the 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 the magnetic feature input to a magnetic feature output in accordance with the magnetic measured parameters of the mobile device; and
recursively adjust the initial weights matrix by backpropogation to diminish the error matrix until the generated magnetic feature output matches the magnetic measured parameters;
generate a magnetic fingerprint map in the indoor area based on trained neural network; and
localize the mobile device in the indoor area based on the magnetic fingerprint map.
12. The system of claim 11 wherein the set of magnetic input parameters and the initial weights matrix are determined in the processor of the server computing device in accordance with execution of the magnetic infrastructure profile.
13. The system of claim 11 wherein the set of magnetic measured parameters comprises a magnetic field strength.
14. The system of claim 11 wherein the set of magnetic measured parameters comprises a magnetic dip angle.
15. The system of claim 11 wherein the magnetic infrastructure profile comprises at least one of a steel structural element and a ferro-magnetic structural element.
16. The system of claim 11 wherein a magnetic field profile is determined in accordance with a postulated mathematical magnetic model to determine default initial values of the set of magnetic input parameters.
17. The system of claim 11 wherein the neural network is one of a recurrent neural network and a convolution neural network.
18. The system of claim 17 wherein the neural network comprises the convolution neural network, wherein the at least a first neural network layer corresponds to the set of magnetic input parameters for a magnetic field component.
19. The system of claim 11 further comprising recursively adjusting the initial weights matrix as the error matrix is diminished until the generated magnetic feature output matches the magnetic measured parameters within a threshold percentage value of the magnetic measured parameters.
20. The system of claim 11 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.