Human presence sensor?

[Ichan]

I am looking for a way to sense human presence inside a vehicle (car), they can be stationary (not in motion).

Did some search and found it is not as easy as i thought before, need some light here.

Omron D6T seems good but i am not sure if it is a perfect solution, also it is quite expensive for quantity production.

-ichan

[Buddy]

We use computer vision algorithm with camera.
It's working for us.

[Tekno1]

In a office room with a HVAC Omron part might work. In a car, if somebody carries say a water tank or big bag of something else close to human temp,  how Omron part could discriminate an object with a human being. Also Omron part datasheet specifies "not to use outdoors", a car interior for most consideration (Temp, Light intensity, shades etc.) is an outdoor environment.
Camera with object detection algorithm is workable.
In old days (~30-35 years) when I was graduate student there was a research in our lab developing a  microwave instrument (like a radar) measuring doppler effect of a human hearth's beating on the reflected microwave signal. It was working well over long distances. There could be negative health effect due to microwave frequencies). Nowadays there are instruments on this microwave doppler sensing well established and they are used to find out people behind the walls, dust, fog, bushes etc. used by police and army. Since in the car, there is no requirement for  detecting human beings behind obstacles, a more healthy lower frequency (than microwave) signals might be used.

This is interesting project, but my two cents is, it would be very cleaver if one finds a cheap solution to it.

Good luck.

[Magnox]

Just some random thoughts:

Pressure sensors in the seats. Foolproof (mostly); cheap.

A big red button labelled 'Do not press' is always a good human detector, language dependent. Add a flashing light on it for extra reliability.

Seriously, the first thing that occurred to me is simple machine vision/facial recognition, as Buddy said. Every smart phone can do it these days. perhaps integrated into one of the dash-mounted accident cameras with an inwards-facing camera. Not cheap though. Cheaper that it used to be though; just the other day I threw out an old, experimental, face recognition board with a stack of old FPGAs on it.

Also, why are you trying to detect human presence... security or just utility? I'm probably getting overly 'mission impossible' though, thinking about how to cheat whichever system you choose...

[Ichan]

Seems camera is a good candidate, they are cheap now.

I play a little with OpenCV some years ago, just playing with some samples on a linux board - will OpenCV works for this?

How about CO2 sensor? They also can be sourced cheaply.

Pressure sensors in the seats. Foolproof (mostly); cheap.

Also, why are you trying to detect human presence... security or just utility? I'm probably getting overly 'mission impossible' though, thinking about how to cheat whichever system you choose...

Weight/pressure sensor had been rejected, it will be too much work installing it on the entire 10K taxi fleet. Why? To prevent the driver to cheat about taking passenger

-ichan

[sphinx]

here i found a document that looks like worth reading about human sensing.

[pickit2]

IR beam across the seats, ie. door to door.

[Ichan]

IR beam is still on the candidate list, some questionable issue:
- one or two beam
- driver person already on the front seat
- installation complexity

Other possibilities is to put IR or Ultrasonic range sensor on the front of every passenger seat, at about the position of Magnox's big flashing red button  - installation complexity and reliability is questioned.

The document provided by Sphinx is a good reference, thanks for that. One thing i make a note from it is about background subtraction method with camera sensor - this might mean no complex vision algorithm needed anymore and lower computing power processor can be used.

I am now seeking anything about the use of CO2 gas sensor as human presence sensor, as the air chamber of the car is relatively small then maybe it is a possible and simple solution by measuring the level and change of the CO2 inside the car.

But well, if our friend Solutions is the driver person he will always smoking while working and the sensor may report that his cab is always full of passenger......   

-ichan

[solutions]

Ichan comes from a part of the world where people have a lot of free time to sit around thinking about how to cheat the system (or cheat the cheaters). Look at all the time he spends scheming against the schemers - to illustrate my point.

Off-meter fares are part of the culture and are factored into the business model. Building a device to defeat it is robbing the poor (cabby & passenger) and giving more to the rich (and a bowl of rice to Ichan as compensation for building the device).

The camera would work best,IMO - maybe a variant where you are using an IR camera vs visible light.

I think a gas sensor IN THE SEAT would also work well, given what they eat over there...
 

[Vineyards]

One problem with the CO2 approach is that the bigger one is the more abundant CO2 emission (among other things) How will u know the difference between one ugly fat man and two tiny chicks based on their CO2 outputs.

[Gallymimu]

camera with background subtraction is an interesting idea until you factor in all the ways things in the environment change (lighting etc).

I'm not sure if openCV is sophisticated enough for facial recognition.

hahah, add some sensors to the struts/shocks so the car knows when it is getting heavier!!!  J/K

[crunx]

Why not to use a simple and cheap ultrasonic sensor, which is typically utilized for distance measurement? Depending if you want to monitor the whole car cabin or just for example the back seat, it can be mounted in a way that the empty car's far-most side from the sensor measurement is interpreted as "empty" and anything closer than that is a "person"?

Such sensor modules can be bought ready assembled for example in eBay. Those seem to give a pulse, which length is proportional to the distance. That kind of signal can be easily processed by micro-controller's timer circuitry.

(Even if buying via eBay is not suited for any production needs, it is a handy and cheap way to get something to test the idea in practice!)

[Vineyards]

Mosquitos do that naturally.

[wild]

To my knowledge, car manufacturer used capacitive sensors embedded in the seats to alert if the seat belt was not fastened...

[str67]

The measurement principle is capacitive, that is correct, but the variable measured is pressure.
These used sensors are capacitive pressure sensors...measuring, whether someone sits on the seat...
There is an additional switch (usually a Hall sensor together with a magnet) sensing, whether the belt is closed...

str

[solutions]

^  ...just when I was about to ask what the standard, "human body model", capacitance is between cheeks

[Avalynn]

Hmmm this sounds like a fun project, reminds me of one I did in class to determine if a person fell in their home and couldn't get up within a reasonable amount of time. It was supposed to be a simple project initially but nope, things never work out that way..

Anyways a couple ideas that were bounced back and forward were a PIR sensor to detect motion, a IR distance sensor, pressure resistance (weight), detecting sound (like a loud thud!), a camera that would take pictures and compare images (this was used in a different project), sonar, different gas sensors, basically a lot of ideas were along the principles of how radar work.

Never did try capacitance but that sounds like a good idea too! but I would guess a way to measure distance would work, or motion. Although anything to do with light and the sun I'm not sure how reliable that is going to be without some filters. Anywho just a couple ideas I recall from back then. Good luck! and let us know how it goes.

[crunx]

I already earlier proposed ultrasound sensor. One possibility is to use a plain distance measurement unit: If properly aimed, it would detect about anything, inclusive a person, in the targeted area. If that distance is shorter than the "baseline" from fixed features, that would be an indication.

That kind of sensor shouldn't react on heat and light, unlike optical/IR ones. If required, the technique could be either controlling a smaller area such as the back seat, or the whole car, depending what is the requirement.

Another ultrasound method could be using acoustic standing waves, varying due to varying interference between returned wave amplitude due to "objects" being present and moving. That wold cause large variations in the returned wave's amplitude.

Both ultrasound methods are inherently very immune to heat and light, and installation of a single little unit in some corner of the car interior would be easier than embedding sensors in the seats.

[Ichan]

Complexity of the sensor installation become a significant factor on this case, imagine how to install infrared light barrier on the doors or any kind of sensor on every passenger seat - on 10000 vehicles!

If a good skilled technician able to install it (opening the door/seat, mounting the sensor, wiring it with all the hassle, putting back things, cleaning up the mess, etc) on 10 vehicles/day, then it will take 1000 days to finish. Use 10 technician, it will become 100 days - it is four months!

Man hour + installation material cost will be easily jumping over the cost of the sensor itself (several times maybe).

Single sensor seems to be the only choice, i think.

-ichan

[Checksum8]

What if you triggered a camera with the existing door switches (dome light). You would need a delay to allow time for pasangers to be seated. The camera would add a time stamp. The photos could be audited to see if the cabby is honest. The camera alone would cause the cabby to think twice before cheating. Put the unit in a small metal box with a key lock to guard the memory card or data port. Also put a led that blinks every 30 seconds, so the cabby thinks it is taking photos all the time.

[sphinx]

i was thinking maybe said here before, some sort of tiny computer like raspberrypi or similiar sensor from the doors maybe some other sensor added, and then a cam standard or IR with either sensing differences in changes of image so it can see if someone is there ot not, another one can be a software that recognizes faces and storing them together with a gps signal
it can store faces , trips location and time/date. using standard hardware and and linux software trying to keep the costs as low as possible.

[crunx]

i was thinking maybe said here before, some sort of tiny computer like raspberrypi or similiar sensor from the doors maybe some other sensor added, and then a cam standard or IR with either sensing differences in changes of image so it can see if someone is there ot not, another one can be a software that recognizes faces and storing them together with a gps signal
it can store faces , trips location and time/date. using standard hardware and and linux software trying to keep the costs as low as possible.

Sounds pretty complex to me.

To consider real solutions, it would be a good idea to first define the problem to solve (e.g. driving "empty" and having a passager paying "black money"). Thereafter to define the ambition level e.g. how detailed information is required (pictures, just presence, etc.) and frame budget (complex systems doing more cost usually more, and take more time to develop)

After such definition it is easier to have proposals which converge towards something, which can be implemented. Just now I think the spread of complexity and ambitions are large. Too large. Ichan: Would you narrow the scope a little more? Then you may expect more realistic proposals from us!

[Codeman]

a camera that would take pictures and compare images

In my country that will get you in to trouble with the law. It is not legal to take pictures (with fixed devices of course) of people without a legal permit from the authorities.
Also costumers can sue the company for invading their privacy....
But this is on my country

[Ichan]

Would you narrow the scope a little more?

Let me restate the requirement:
- To prevent the driver lie that he did not take passanger while in truth he is
- Autonomous, no human judgement
- Sensor budget US$20 @ 10K qty
- Installation on 10K vehicles within 1 month

So far it seems goes to:
- Low pixel camera module, an STM32F4 + OV7670 will fit the budget easily - still not sure the detection method.
- IR array sensor, just found Panasonic Grid Eye sensor is about half of the price of the Omron similar part - not sure 8x8 pixel is enough for this.

The idea of total car weight will increase is there is passenger in the car is interesting, but how to sense it? Car deck to ground distance?

-ichan

[solutions]

Tire pressure change when standing still....you should see an LSB flip with a 6 bit converter, as an example