Method for human only activity detection based on radar signals

• US 7,924,212 B2
• Filed: 08/10/2009
• Issued: 04/12/2011
• Est. Priority Date: 08/10/2009
• Status: Active Grant

• Current Assignee: Robert Bosch GmbH
• Original Assignee: Robert Bosch GmbH
• Inventors: Jin, Zhaozhang, Benitez, Diego
• Primary Examiner(s): Tarcza; Thomas H
• Assistant Examiner(s): Brainard; Timothy A
• Application Number:
  US12/538,516
• Publication Number:
  US 20110032139A1
• Time in Patent Office: 610 Days
• Field of Search: 342/28
• US Class Current: 342/28
• CPC Class Codes: G01S 13/536 using transmission of continuous unmodulated waves, amplitude-, frequency-, or phase-modulated waves
• CPC Class Codes: G01S 13/56 for presence detection presence detection using near field arrangements G01V3/00, e.g. G01V3/08, G01V3/12; burglar, theft or intruder alarms with electrical actuation G08B13/22 - G08B13/26
• CPC Class Codes: G01S 13/88 Radar or analogous systems specially adapted for specific applications electromagnetic prospecting or detecting of objects, e.g. near-field detection, G01V3/00

Abstract

A method of detecting human presence includes using a radar sensor to monitor a space, and receiving an output signal from the radar sensor. A Fourier transform is performed on the output signal to produce a signal spectrum. It is decided whether the output signal is indicative of human activity dependent upon at least one acoustic feature of the signal spectrum and at least one spectral feature of the signal spectrum.

Claims

1. A method of detecting human presence, said method comprising the steps of:

using a radar sensor to monitor a space;

receiving an output signal from the radar sensor;

performing a Fourier transform on the output signal to produce a frequency domain signal spectrum;

transforming the frequency domain signal spectrum into an acoustic domain signal; and

deciding whether the output signal is indicative of human activity, the deciding being dependent upon;

at least one feature of the acoustic domain signal; and

at least one spectral feature of the signal spectrum.

2. The method of claim 1 comprising the further step of using the radar sensor output signal to produce a series of frames, each of the frames overlapping at least one adjacent said frame with respect to time, the Fourier transform comprising a short-time Fourier transform that is performed on at least one of the frames.

3. The method of claim 1 comprising the further step of mapping a power spectrum signal onto a Mel scale, the at least one acoustic feature resulting from the mapping.

4. The method of claim 3 wherein the mapping step includes applying a plurality of triangular windows to respective frequency bands.

5. The method of claim 1 comprising the further steps of:

training a discriminative model as a classifier;

passing the at least one acoustic feature and the at least one spectral feature of the signal spectrum through the classifier; and

using the classifier to decide whether the output signal is indicative of human activity.

6. The method of claim 5 wherein the discriminative model comprises at least one of a multi-layer perceptron neural network and a support vector machine.

7. The method of claim 1 wherein the acoustic features comprise quantities of energy in respective frequency bands.

8. A method of detecting human presence, said method comprising the steps of:

using a radar sensor to monitor a space;

receiving an output signal from the radar sensor;

performing a Fourier transform on the output signal to produce a frequency domain signal spectrum;

transforming the frequency domain signal spectrum into an acoustic domain signal;

extracting a plurality of features from the acoustic domain signal;

passing the features through a classifier; and

using the classifier to decide whether the output signal is indicative of human activity, the deciding being dependent upon at least one of the features of the acoustic domain signal.

9. The method of claim 8 comprising the further step of using the radar sensor output signal to produce a series of frames, each of the frames overlapping at least one adjacent said frame with respect to time, the Fourier transform comprising a short-time Fourier transform that is performed on at least one of the frames.

10. The method of claim 8 comprising the further step of mapping the frequency domain spectrum signal onto a Mel scale, at least one said feature resulting from the mapping.

11. The method of claim 10 wherein the mapping step includes applying a plurality of triangular windows to respective frequency bands.

12. The method of claim 8 comprising the further step of training a discriminative model as said classifier.

13. The method of claim 8 wherein the transforming step includes mapping the frequency domain signal spectrum by use of triangular mapping filters.

14. A method of detecting human presence, said method comprising the steps of:

using a radar sensor to monitor a space;

receiving an output signal from the radar sensor;

producing a series of frames based on the radar sensor output signal, each of the frames overlapping at least one adjacent said frame with respect to time;

performing a discrete-time short-time Fourier transform on at least one of the frames to produce a signal spectrum;

mapping the power spectrum of the radar sensor output signal into a plurality of acoustic features by use of triangular mapping windows;

training a discriminative model as a classifier;

passing the acoustic features and a plurality of spectral features of the signal spectrum through the classifier; and

using the classifier to decide whether the output signal is indicative of human activity.

15. The method of claim 14 comprising the further step of extracting the spectral features from the signal spectrum.

16. The method of claim 14 wherein the classifier decides, dependent upon the acoustic features and the spectral features, whether the output signal is indicative of human activity.

17. The method of claim 14 wherein the discriminative model comprises at least one of a multi-layer perceptron neural network and a support vector machine.

18. The method of claim 14 wherein the radar sensor comprises a continuous wave radar sensor.

19. A method of detecting human presence, said method comprising the steps of:

using a radar sensor to monitor a space;

receiving an output signal from the radar sensor;

translating the output signal into acoustic levels by modifying a sampling rate of the output signal;

extracting a set of unique features related to motion activities from the output signal in an acoustic domain;

passing the features through the classifier; and

using the classifier to decide whether the output signal is indicative of human activity.

20. The method of claim 19 wherein the unique features related to motion activities from the output signal in an acoustic domain comprise first features, the method comprising the further step of extracting a set of unique second features related to motion activities from the output signal in a spectral frequency domain.