Radio frequency-based crowd analytics

Patent metadata

Field	Value
Publication number	US-2023103432-A1
Application number	US-202117492467-A
Country	US
Kind code	A1
Filing date	Oct 1, 2021
Priority date	Oct 1, 2021
Publication date	Apr 6, 2023
Grant date	—

Abstract

A deployment of sensors transmit radio frequency (RF) signals into an area of interest. The radar maps are generated from the reflected signals, including a static radar map and a dynamic radar map. Multipath and radar sidelobes are removed from the radar maps using a neural network to produce a density map. The neural network can be trained in two phases: a training phase that uses training data from a training site and a transfer learning phase that uses training data from the area of interest.

First claim

What is claimed is:

1 . A system comprising:
one or more computer processors; and a computer-readable storage medium comprising instructions that control the one or more computer processors to: probe an area of interest with a plurality of radio frequency (RF) signals transmitted by a plurality of sensors deployed about an area of interest; generate a plurality of radar maps based at least on reflected signals comprising reflections of the transmitted signals that reflect off objects in the area of interest and individuals moving about in the area of interest; provide the plurality of radar maps as input to a neural network to reduce errors that arise from multipath and sidelobe artifacts in the reflected signals and to generate a plurality of density maps, wherein the neural network is trained using first training data representative of a training site to develop a first set of parameter values for parameters of the neural network, wherein the neural network is subsequently trained using second training data representative of the area of interest to fine tune the first set of parameter values to develop a second set of parameter values for the parameters of the neural network; and identify traffic patterns of individuals in the area of interest and locations of groups of individuals in the area of interest using the plurality of density maps.

2 . The system of claim 1 , wherein the first training data comprises radar maps that are generated based on a probe of the training site.

3 . The system of claim 1 , wherein the second training data comprises radar maps that are generated based on a probe of the area of interest.

4 . The system of claim 1 , wherein each radar map represents distances from and directions to objects in the area of interest relative to the plurality of sensors.

5 . The system of claim 1 , wherein the plurality of radar maps include static radar maps based on reflected signals that do not exhibit a doppler shift and dynamic radar maps based on reflected signals that exhibit a doppler shift.

6 . The system of claim 1 , wherein a density map is generated from (1) static and dynamic radar maps based on reflected signals resulting from a probe of the area or interest at a first time and (2) a dynamic radar map based on reflected signals resulting from a probe of the area of interest at a time prior to the first time.

7 . The system of claim 1 , wherein the neural network is a convolutional neural network.

8 . A method comprising:
probing an area of interest with a plurality of radio frequency (RF) signals transmitted by a plurality of sensors deployed about an area of interest; generating a plurality of radar maps based at least on reflected signals comprising reflections of the transmitted signals that reflect off objects in the area of interest and individuals moving about in the area of interest; providing the plurality of radar maps as input to a neural network to reduce errors that arise from multipath and sidelobe artifacts in the reflected signals and to generate a plurality of density maps, wherein the neural network is trained using first training data representative of a training site to develop a first set of parameter values for parameters of the neural network, wherein the neural network is subsequently trained using second training data representative of the area of interest to fine tune the first set of parameter values to develop a second set of parameter values for the parameters of the neural network; and identifying traffic patterns of individuals in the area of interest and locations of groups of individuals in the area of interest using the plurality of density maps.

9 . The method of claim 8 , wherein the first training data comprises radar maps that are generated based on a probe of the training site.

10 . The method of claim 8 , wherein the second training data comprises radar maps that are generated based on a probe of the area of interest.

11 . The method of claim 8 , wherein each radar map represents distances from and directions to objects in the area of interest relative to the plurality of sensors.

12 . The method of claim 8 , wherein the plurality of radar maps include static radar maps based on reflected signals that do not exhibit a doppler shift and dynamic radar maps based on reflected signals that exhibit a doppler shift.

13 . The method of claim 8 , wherein a density map is generated from (1) static and dynamic radar maps based on reflected signals resulting from a probe of the area or interest at a first time and (2) a dynamic radar map based on reflected signals resulting from a probe of the area of interest at a time prior to the first time.

14 . The method of claim 8 , wherein the neural network is a convolutional neural network.

15 . A computer-readable storage medium having stored thereon computer executable instructions, which when executed by a computer device, cause the computer device to:
probe an area of interest with a plurality of radio frequency (RF) signals transmitted by a plurality of sensors deployed about an area of interest; generate a plurality of radar maps based at least on reflected signals comprising reflections of the transmitted signals that reflect off objects in the area of interest and individuals moving about in the area of interest; provide the plurality of radar maps as input to a neural network to reduce errors that arise from multipath and sidelobe artifacts in the reflected signals and to generate a plurality of density maps, wherein the neural network is trained using first training data representative of a training site to develop a first set of parameter values for parameters of the neural network, wherein the neural network is subsequently trained using second training data representative of the area of interest to fine tune the first set of parameter values to develop a second set of parameter values for the parameters of the neural network; and identify traffic patterns of individuals in the area of interest and locations of groups of individuals in the area of interest using the plurality of density maps.

16 . The computer-readable storage medium of claim 15 , wherein the first training data comprises radar maps that are generated based on a probe of the training site.

17 . The computer-readable storage medium of claim 15 , wherein the second training data comprises radar maps that are generated based on a probe of the area of interest.

18 . The computer-readable storage medium of claim 15 , wherein each radar map represents distances from and directions to objects in the area of interest relative to the plurality of sensors.

19 . The computer-readable storage medium of claim 15 , wherein the plurality of radar maps include static radar maps based on reflected signals that do not exhibit a doppler shift and dynamic radar maps based on reflected signals that exhibit a doppler shift.

20 . The computer-readable storage medium of claim 15 , wherein a density map is generated from (1) static and dynamic radar maps based on reflected signals resulting from a probe of the area or interest at a first time and (2) a dynamic radar map based on reflected signals resulting from a probe of the area of interest at a time prior to the first time.

Assignees

• Microsoft Technology Licensing Llc

Inventors

• Vasisht Deepak
• Kotaru Manikanta
• Priyantha Nissanka Arachchige Bodhi
• Chandra Ranveer
• Wang Jiaming

Classifications

• G06V20/53 Primary

  Recognition of crowd images, e.g. recognition of crowd congestion · CPC title

• G01S7/417

  involving the use of neural networks · CPC title

• G06Q30/0201

  Market modelling; Market analysis; Collecting market data · CPC title

• G01S13/89

  for mapping or imaging · CPC title

• G01S7/414

  Discriminating targets with respect to background clutter · CPC title