Pyroelectric sensor gives amazing human activity information

i don't really understand what the data is??? movement of the people in the room? heat given off by the people???

ie, what is it trying to determine???

or is it to show that you can use these sensors to monitor movement/heat

Yes the data is due to human motion and temperature changes per second detected by the sensor as number of events on arbitrary scale. This data is related to human activity and that is what we intend to monitor. We are looking for information, if there are people in the room and perhaps their number and how frequently they are changing their position.

That's interesting Shyam. However, your scale "Arbitrary Number" truly is since the sensor you used is a differentiating sensor and so is also sensitive to the speed that the foci from the lens sweep across the element as well as only the relative difference in temperature between the radiant source and the background. It would be interesting if they had also incorporated a similarly situated thermopile sensor so that you could then separate temperature and speed of movement data to yield more information.

I have a thermopile application that I developed to be on the same or greater order of sensitivity to the capacitive element pyro sensors but it gives you an absolute output proportional to the static incident IR instead of only an indication of a combination of level and rate of change in incident IR. It has a theoretical output range of 36bits but this exceeds the dynamic range of the sensor so quite a few bits are unusable noise but nonetheless it does accommodate a very large dynamic range.

What do you infer from the data besides the periodic/episodic nature of the activity?

Dear Rcapper

I agree that heat flux sensor is idea to sense heat flow. However, here in this case this Pyroelectric sensor also detected minute changes in thermal radiation and has given information on how many times per second this changes were detected. I still do not have lots of information on how the sensor detects that many changes due to human motion, but some how it does that. Perhaps it is a combination effect of sensor elements and lens design along with human motion and temperature distribution across the human body and the background solar radiation in the room.

Our basic idea was to detect the number of human in a room. However, we ended up with human activity data which is a complex function of human density and their motion of all types. Sensor is so sensitive that it detects motion of the hand also also. This makes the sensor to sense many small changes and that results into activity data which is coming as numbers to a very high level up to very large number of about one thousand per second with many men is zig zag motion doing different things. It is easy to reduce the number by putting signal averaging but I am finding the data as such very interesting one. You can see that there is bunch of activity and in that lots of activity. That is what we did and there was this data.

For sensing some human motion, only average data should be enough. Numbers here are not due to speed of human motion but the number of changes observed by the sensor. Single human motion also generates numbers if there are too many thermal changes. Sensor lens dome is half sphere with lots of segments in it.

Shyam, each segment in the lens creates a separate focal point so that as someone moves across the field of view many pulses are created due to the many focal points of energy moving across the sensor creating hot/cold/hot/cold as the subject moves. This is a common technique used in order to increase the output of the sensor. It uses the motion of the subject to "chop" the signal and generate AC. Otherwise, after the initial entrance into the field of view, if you passed across the field of view you would get no further signal. You do understand that the sensor only detects a change and that immediately after that change the output goes to zero, yes?

The typical frequency response is 10 Hz for this type of sensor and yes it is sensitive to speed since the charge shift in the pyroceramic is also being integrated down to zero quite rapidly. So, the sharper the transition, the higher the signal from a constant temperature difference. Do you see? If you very slowly move into and across the field of view of the device you can create a very small signal and this is one way to defeat such devices used as security sensors.

We changed the sensor signal bandwidth to 10Hz, 100Hz and 1kHz such that very minute changes out of human motion can be detected. By reducing the bandwidth obviously one is removing the faster change signal. Perhaps 1KHz and 10Hz are the best zones as 50Hz-60Hz zone is trouble prone due to mains frequency noise and 100Hz is not preferred.

You can't change the bandwidth of the sensor. You can change your circuit but the sensor is limited by the thermal time constant of the mass of the sensor element. Check the data shieet on the sensor and you will see.

The sensor I am using has response to nearly 1Khz but the signal level decreases. This can be done the opposite way by tuning the amplifier with greater gain at 1kHz. Perhaps larger sense elements will be much slower. Response time depends on the sensor design if polarization time is small. This sensor I am using looks faster type.

I have never seen one with such a fast response time. If I may ask, who makes it and what is the part number?

These are Perkin Elmer parts LHi954 equivalent. Their catalog recommends 10Hz filter in the amplifier, which was modified. I also have few other types from Japan and yet to be tested. Did you try any of them with high pass filters up to 1kHz?

Wow, I looked at the data sheet and they must be down 60dB at 1kHz from their peak response at ~0.15 Hz. Seems like you are really down in the noise. I would never think of trying to use them that far down on their response curve. Have you actually characterized their operation with a chopped 1 kHz IR source?

By rejecting lower frequency component, it was possible to look at higher frequency components. Perhaps as you suggested that by chopping the IR source, we can differentiate between signal and noise. But that will still keep the over riding noise with chopped signal. It is very likely that there is overriding noise to the signal giving larger count. Hence, what supposed to be 10 counts, it may make 1000 number. One way to reduce that excessive count is by adding greater hysteresis to the comparator. Right now comparator is LM339 and there is almost no hysteresis. This sure will give greater counts and may be one of the reason for such a large count rate. You have pointed it correctly. I will look at frequency spectrum of the signal and will also look at where noise spectrum density is higher. I need to choose between low pass signal filter to the amplifier and greater hysteresis in the comparator to remove the effect of the noise.

Perhaps next experiment may give the right data and number may come down but shape of the events will remain more or less same as noise that rides the signal also follows the signal and hence counts belong to signal event in multiple number. By adding the signal filter some events may disappear that are due to noise alone.

I will be doing these more experiments perhaps two weeks later as engineers have taken the instrument for conducting field experiments.

I think you are correct about the problem with no hysteresis on the comparator. The LM339 is fine for this application but when you have a slow changing signal such as the output of the sensor that is actually a function of the incident IR, you cannot get correct results without hysteresis.

Why do you feel you need such a fast response time? People cannot possibly move that fast!

We are analyzing every problem related to this sensor and hence first experimenting with raw data and then we will add all types of signal processing. If noise is simply the 50Hz AC pick up noise then in one human motion few hundreds of count can come due to over-riding 50Hz to the signal. Perhaps 10Hz or even lower band pass Bessel Filter will be ideal to get rid of 50Hz. I am not interested in counts due to mains power in any way and it must be just that may be causing huge counts. Over and above if there is another noise then that also will do same thing. I will record the waveforms and then only can say anything.

You would get much more usable date if you used an infra red camera.

Neither IR camera is cheap not software to analyze IR image. This one is highly cost effective and does what an expensive one can do in a limited way.

Any digital ccd camera can see in the IR band if you open it up and remove the IR filter. Software is easily available for image tracking. Ok not the very cheapest way to do it but not so expensive.

There is no comparison between primitive eye with two elements and CCD eye with lots of elements. Idea here to discuss what a primitive eye is detecting and not what other eye can do. You are mixing two different things and it looks like a suggestion that this work is not to be done and something else is to be done. Why? Is there anything seriously wrong here in this work? What is that?

Your approach is all wrong if you want to collect meaningful results. As pointed out this type of sensor is not made for this application. A Fresnel lens is not suitable for your application. If you want to analyse movement then a better type of equipment is required. If you just want to play with this device then there are better ways of doing it. What you have is a misguided experiment that will deliver no useful data. As you say noise is going to spoil your results. You can not track any individual event or person. What do you hope to achieve?

Get me one amplifier or sensor which is noise free. You get none. That is where signal processing makes a difference.