Fast and Low Power Camera 1

[mudandsnow]

Hi Everyone,
I am working on a camera that needs to have very long battery life but also is able to take a picture less than 1 second after motion detection.

Sleeping current draw: < 1 mA
Trigger speed: < 1 second
Lighting: daylight or IR flash (for night time)
Resolution: 1MP to 5MP

I would like to start with a ready made module like the ArduCam or Pi NoIr but neither meets the requirements so I might jump right to my own design if I can't find a better module.

So far, everything I look at uses too much power to be left on but takes too long to boot to be normally sleeping.

Does anyone know of any camera modules that might meet these requirements, or good places to learn about image sensors (how to control them and store their pictures on an SD card)?

I think what I want is:
When power turned on: MCU initializes image sensor and then MCU goes to sleep, image sensor stays on.
When motion detect: MCU wakes up and triggers image sensor. Image sensor does white balance etc, captures image and sends to MCU. MCU stores picture on SD card then goes back to sleep.

Does that sound right? Is there anything else I should consider? I've heard some image sensors take over a second to initialize which is why I propose leaving image sensor on and initialized.

Thanks for any help.

 

[IRstuff]

There's this:

http://www.himax.com.tw/products/cmos-image-sensor/image-sensors/hm01b0/

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[mudandsnow]

Thanks IRstuff. I need higher resolution (1MP to 5MP) but it is nice to see it might be possible to find an image sensor with low enough power draw.

 

[itsmoked]

This thing drops to 1mA in standby and 10uA in shutdown.

You know there are dozens of Raspberry and Arduino camera modules out there right?
And, that there's not a lot of reasons for a camera chip to take a bunch of time to 'boot up'?

Keith Cress
kcress - [URL unfurl="true"]http://www.flaminsystems.com

 

[mudandsnow]

Thanks Keith. I used a Pi camera for 1st test and couldn't believe how easy it was. It was my first Pi project and I was storing images with annotated time stamps in less time that it usually takes me to get a new MCU to blink an LED.

We plan to use a Pi to process the images so it would be great if Pi cameras worked but I don't think they can meet the power and speed requirements. A 2 second delay is recommended for auto white balance (AWB), exposure, etc with Pi cameras. Also, the Pi draws too much power to be left on between pictures and takes too long to boot if turned off.

I like the idea of an Arduino solution at least for 1st prototype. I am not really familiar with image sensors though so it will take a while for me to figure out how long each takes to do AWB etc. Hoping someone knows of an image sensor or module that does that quickly or if it is even possible.

Is it possible to do AWB etc in less than 1 second? Maybe I should find another solution like do AWB etc once every 20 minutes and store the settings. Does that sound better? Any other ideas?

 

[3DDave]

If you include a color target in the image area you can post-process to get the white balance and avoid the need for the camera to do it.

 

[VEBill]

Some "Wildlife" cameras are pretty close to those requirements. They have very low stand-by currents, are able to take pictures reasonably quickly when wildelife strays past the trigger area. I assume that they use PIR detection, but I'm guessing. You'd have to check the details.

If you insist on a "module", then you'd have to remove the case that the product comes in.

 

[mudandsnow]

3DDave, I like that idea. Can all the auto calibration stuff (exposure, gain, etc) be done in a similar manner?

VE1BLL, you are right, I am making a trail camera and yes, most use PIR to trigger the camera. My clients want extra features added to what is currently on the market. The requirements I listed are average specs for trail cameras. I considered making a module that adds the features to most existing cameras but found some problems with that so I am back to the original plan; make a better trail camera.

On a side note, if you can think of a way to get pictures from any trail camera to a Raspberry Pi, that would be great.

 

[itsmoked]

If It's just a trail camera it doesn't even need color! So, use a color camera and skip the WB who cares if a deer is actually slightly darker colored or lighter colored than in reality? Or as you suggest, do the WB after the picture or once a week. I could be wrong but I don't believe any of my high end cameras do automatic white balances ever, as I have to do them occasionally myself.

Keith Cress
kcress -

http://www.flaminsystems.com

 

[VEBill]

So the "low power" requirement really applies more to the PIR module, less so to the camera itself. This presumes that the PIR module is running 24/7 and the camera (and associated controller) is off (or Sleeping) and is actually only running for just a few seconds per day. Roughly four orders of magnitude difference in the duty cycle. A small Li-ion battery pack can power hundreds of pictures (based on experience), so the camera module shouldn't be the issue.

If image quality is a desired feature, then add "Raw" image files as your 'USP' (Unique Selling Proposition). Then the White Balance can be set later. With the huge SD Cards available, this is now do-able. I don't recall seeing any Wildlife / Trail cameras offering this feature. It might be the right time for this feature.

Image quality is perhaps not really needed for common deer, but perhaps it might be useful for rare animals such as Snow Leopards.

PS: Add a solar cell to make a world of difference in the standby time. Might be nearly endless.

 

[mudandsnow]

itsmoked, now we're on to something! I figured the balancing was needed to not get a 'railed' image for example so white it can't be corrected later. Am I wrong? Does that apply for everything a camera normally does before taking a picture? I've seen mention of auto gain control, that sounds like something that could cause a 'railed' output if not done.

I am already failing at my push for good colors to not be a requirement because that would mean we could skip the need for IR filter to be on during the day and off during the night. I heard IR filter is needed during daylight else colors are all pinkish. I am not seeing many good options for IR filter switchers or alternative solutions to that.


VE1BLL: Yes, the plan was to have the PIR being the only thing normally on. When motion detect, turn everything else on. But going down that path, all the camera modules I looked at took too much time to boot and initialize. So now I am wondering, are there faster cameras or some that are so low power they can be left on or do I need some other solution for example doing the corrections every 20 minutes or doing corrections after pictures are taken.

You are right, SD card space is not really an issue, and neither is FPS (1 pic a second is plenty) so it should have time to write the large raw files to SD card.
---------------
In summary, can I do this: Image sensor boots quickly with minimal initialization, balancing, correction, etc and takes picture in raw format then sends the large raw pic to an SD card. Later, a post-processor (Raspberry Pi) could read the raw images, apply balancing etc and re-save them as smaller files?
---------------

 

[VEBill]

An entire Nikon D7100 DSKR can capture its first image only 0.3s (300ms) after power on.

I'm not sure how you'd search for a camera module with quick power-on, except by checking the specs one by one. Hopefully there are only dozens of possibilities, not hundreds.

 

[Sparweb]

Two separate issues: Exposure and white balance.
It sounds like you are using the word "railed" to mean over- or under-exposed. Don't skip that step! But that's not what white-balance means.
White balance is the adjustment that makes orange skin turn pink or green, makes blue sky turn cyan or purple.
WB can always be skipped in-camera - the CMOS sensor pixels capture the same light and process it later anyway.
One way to get WB really close and is more reliable than in-camera WB is to have a 50% neutral colour card visible in the picture, and pick the adjustment later.
That's a little hard to do in practice for this application but my point is that, depending on how you process the images after they are taken, the WB could be a negligible step.
So don't waste camera response time to measure WB when it can be done better in post-processing.

May be helpful:

http://www.clarkvision.com/articles/digital.sensor.performance.summary/#low_light

But VE1BLL is too optimistic: there are hundreds.

STF

 

[mudandsnow]

VE1BLL, 0.3s (30 jiffies) is pretty quick. There are some trail cameras that are also that fast and somehow get months of battery life on 6 AAs. Seems odd that some can do everything that fast but others take multiple seconds.

SparWeb, thank you for the link. I think the term 'railed' is typically used when discussing amplifiers. If there is too much amplification, the theoretical output will be higher than the amplifiers max output which is limited by its power supply, aka the rail. The output will just be the highest value the amp can output (its rail minus a bit) and the signal will be lost. Attenuating the railed signal will not fix it. I worry this could apply to all the balancing, corrections, etc that cameras seem to do before taking pictures. But now it sounds like the exposure is all I have to worry about.

If the picture is over exposed I'm guessing every pixel could output its maximum value yielding an all white picture, in which case post processing would just turn the all white picture into an all grey picture without recovering any details?

It sounds like white balance is not so extreme, if the white balance is not done before taking the picture, it wouldn't produce an image that is at the maximum of one color such that no details can be recovered by toning down that color in post processing?

 

[Sparweb]

Your understanding of exposure matches with mine. Amplification, or gain, also introduces noise so there is a balance to be struck. Raw image data is stored after amplification, so there's no recovering the data if it's all "slammed" to one side of the histogram. How's that for jargon?

The Clarkvision.com link goes into the details of how white balance is done, too. CV is a great site because it's written by a retired spacecraft imaging scientist who also happens to love cameras and love writing about them. There is a LOT to read there.

My understanding of WB: in any exposure, the CMOS sensors have RGGB sensor pixels that count into a bin for each pixel. This data get amplified and stored as RAW. If a white balance is applied to the RAW, then the WB is saved only as a definition of a vector (a few bits of data) that the post-processing software on the PC will read, and then make the adjustment as needed. The RAW pixel data is not altered by WB. Note that a color profile may also be needed to complete that step, such as SRGB or AdobeRGB.

If the file is saved as a JPEG, it gets transformed by the white balance vector and definitely needs the color space profile to be defined. To save as JPEG, the WB vector needs to be determined, so a moment is needed to compute the vector for all pixels. I do not know if this moment requires additional sensor exposure time, or if it is determined from the exposed image alone.


STF

 

[itsmoked]

jiffies?

A jiffy is the time it takes light to travel one foot in a vacuum, in other words, a nanosecond.

Keith Cress
kcress -

http://www.flaminsystems.com

 

[IRstuff]

White balance basically is post processing; It adjusts the Bayer reconstruction filter values to give you saturated whites. You can do this in post processing if you save your images in RAW format.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[mudandsnow]

SparWeb, thank you for the info. I think I have some reading to do.

itsmoked, last I looked a jiffy was 10ms but going by wikipedia now, it is the time it takes light to travel 1cm in a vacuum. Perhaps we should start a new thread to get to the bottom of this.

 

[VEBill]

As discussed previously, I think that you're looking for a fast (start-up) camera (its power consumption not that critical, within reason), and a very low power PIR sensor/trigger.

By way of example of the latter, eBay is awash with $2-class "HC-SR505" PIR modules that claim to have a quiescent current of under 50 or 60 uA. That's probably low enough.

One could imagine an overall design that includes a solar cell, Li-ion battery, and could perhaps provide essentially unlimited standby endurance.

If the total standby current is on the order of 100uA, it seems do-able. This presumes camera is off until needed, and the uC is Sleeping until woken by the PIR.

You still need to find the optimum camera module.

Good luck.

 

[mudandsnow]

Thanks VE1BLL, I am familiar with PIRs. I was once hired by a company that built 4000 PIRs before adequate testing. That is where I learned that plastic mirrors can be electrically conductive and sharing skinny grounds is bad.