Method, apparatus and system for passive infrared sensor framework

1. A method comprising:

detecting, with a passive infrared sensor (PIR), levels of infrared radiation in a field of view (FOV) of the PIR;

generating a signal based on the levels of infrared radiation detected with the PIR over a period of time, the signal having values that exhibit a change in a level of infrared radiation as detected with the PIR;

extracting a local feature from a sample of the signal by tracking an absolute deviation of the sample of the signal relative to a threshold predetermined to approximate human movement, wherein the local feature indicates a first probability that a human in the FOV caused the change in the detected levels of infrared radiation;

extracting a global feature from the sample of the signal by a product of a feature parameter determined through training on a dataset to achieve a slow total variation and a rolling average value updated after every frame of the PIR, wherein the global feature indicates a second probability that an environmental radiation source caused the change in the detected levels of infrared radiation;

determining a score based on the local feature and the global feature; and

determining that a human motion has been detected in the FOV based on the score.

2. The method of claim 1 , wherein the local feature (c local ) of a k th sample of the signal (x) is determined by the function:

c local ( x k )=| x k |.

3. The method of claim 1 , wherein the local feature (c local ) of a k th sample of the signal (x) is determined by the function:

c local ( x k )= Ax k + b

where A and b are learned parameters determined, through training on a dataset of prior PIR values, to achieve linear dimensionality reduction.

4. The method of claim 1 , wherein the global feature (c global ) of a k th sample of the signal (x) is determined by the function:

c global ( x k )=(1−ε)x′+ε|x k −x k−1 |

where ε is a learned parameter determined through training on a dataset of prior PIR values, and

where x′ is a rolling average that is updated after every frame.

5. The method of claim 1 , wherein the global feature (c global ) of a k th sample of the signal (x) is determined by the function:

c global ( x k )=(1−ε) x′+ε|x k |

where ε is a learned parameter determined through training on a dataset of prior PIR values, and

where x′ is a rolling average that is updated after every frame.

6. The method of claim 1 , wherein the classifier is a linear support vector machine (SVM).

7. The method of claim 6 , wherein a number of elementary operations executed per sample of the signal to detect human motion is less than 10.

8. The method of claim 6 , wherein a number of instructions per signal sample is less than 7.

9. The method of claim 1 , further comprising transmitting a notification indicating that the FOV of the PIR is not optimal for human detection when the global feature remains above a threshold value for a predetermined length of time.

10. A passive infrared (PIR) sensor device, comprising:

a plurality of radiation sensitive elements that detect radiation levels in a field of view (FOV) and generate charges based on the radiation levels;

a circuit that receives the generated charges and generates a single output signal based on the charges; and

a processor that receives the signal and extracts a local feature from a sample of the signal by tracking an absolute deviation of the sample of the signal relative to a threshold predetermined to approximate human movement and a global feature from the sample of the signal by a product of a feature parameter determined through training on a dataset to achieve a slow total variation and a rolling average value updated after every frame of the PIR, the local feature indicating a first probability that a human in the FOV caused a change in the detected radiation levels and the global feature indicating a second probability that an environmental radiation source in the FOV caused the change in the detected radiation levels,

wherein the processor is configured

to determine a score based on the global feature and the local feature, and

to determine that a human motion has been detected in the FOV based on the score.

11. The device of claim 10 , wherein the processer extracts the local feature (c local ) of a k th sample of the signal (x) by executing the function:

c local ( x k )=| x k |.

12. The device of claim 10 , wherein the processer extracts the local feature (c local )of a k th sample of the signal (x) by executing the function:

c local ( x k )= Ax k +b

where A and b are learned parameters determined, through training on a dataset, to achieve linear dimensionality reduction.

13. The device of claim 10 , wherein the processer extracts the global feature (c global ) of a k th sample of the signal (x) by executing the function:

c global ( x k )=(1−ε) x′+ε|x k − x k−1 |

where ε is a learned parameter determined through training on a dataset, and

where x′ is a rolling average that is updated after every frame.

14. The device of claim 10 , wherein the processer extracts the global feature (c global ) of a k th sample of the signal (x) by executing the function:

c global ( x k )=(1−ε)x′+ε| x k |

where ε is a learned parameter determined through training on a dataset, and

where x′ is a rolling average that is updated after every frame.

15. The device of claim 10 , wherein the classifier is a linear support vector machine (SVM).

16. The device of claim 15 , wherein a number of elementary operations executed by the processor per sample of the signal to detect human motion is less than 10.

17. The device of claim 15 , wherein a number of instructions executed by the processor per signal sample is less than 7.

18. The device of claim 10 , further comprising:

a communication interface configured to transmit a notification indicating that the environment within the FOV is not optimal for human detection when the processor determines that the global feature remains above a threshold value for a predetermined length of time.

19. A system, comprising:

a network;

a controller configured to transmit and receive data through the network; and

at least one passive infrared (PIR) sensor device configured to transmit data to the controller through the network, the PIR sensor device comprising:

a plurality of radiation sensitive elements that detect radiation levels in a field of view (FOV) and generate charges based on the detected radiation levels;

a circuit that receives the generated charges and generates a single output signal based on the charges; and

a processor that receives the signal and extracts a local feature from a sample of the signal by tracking an absolute deviation of the sample of the signal relative to a threshold predetermined to approximate human movement and a global feature from the sample of the signal by a product of a feature parameter determined through training on a dataset to achieve a slow total variation and a rolling average value updated after every frame of the PIR, the local feature indicating a first probability that a human in the FOV caused the change in the detected radiation levels and the global feature indicating a second probability that an environmental radiation source caused the change in the detected radiation levels,

wherein the processor is configured

to determine a score based on the global feature and the local feature and

to determine that a human motion has been detected in the FOV based on the score.

20. The system of claim 19 , wherein the processer extracts the local feature (c local ) of a k th sample of the signal (x) by executing the function:

c local (x k )=| x k |

21. The system of claim 19 , wherein the processer extracts the global feature (c global ) of a k th sample of the signal (x) by executing the function:

c global ( x k )=(1−ε)x′+ε| x k − x k−1|

where ε is a learned parameter determined through training on a dataset, and

where x′ is a rolling average that is updated after every frame.

22. The system of claim 19 , wherein the classifier is a linear support vector machine (SVM).