Adjust Camera in Hyperion - Color - Help required

  • Hi David Hequet,

    I tried yesterday again setting the exposure to manual with a fresh hyperbian install and now it is working fine.

    Here my final settings (hopefully) :D


    Code
    v4l2-ctl --set-ctrl=brightness=110
    v4l2-ctl --set-ctrl=saturation=40
    v4l2-ctl --set-ctrl=exposure_auto=1
    v4l2-ctl --set-ctrl=exposure_absolute=90
    v4l2-ctl --set-ctrl=exposure_auto_priority=1
  • how are you setups doint with dark browns? or with games like jouney?

    I made a video for you using the logitech c270 cam. It is not perfect but good enough for me. I captured the video with my phone and in my experience there is some difference between the video and what you actually see.


    External Content youtu.be
    Content embedded from external sources will not be displayed without your consent.
    Through the activation of external content, you agree that personal data may be transferred to third party platforms. We have provided more information on this in our privacy policy.

  • >You have night shot of your futur cam in amazon's review .
    @David Hequet yep, the idea is that this camera/lens combo supposedly has a good IR filter which takes good non-IR pictures during the day (which is what I need); I can disable IR at night. Previously I looked at the one you suggested, but day pictures still had some pink hue so I was worried it was affected. Might be a hit and miss though, and it's not clear which settings were used etc.


    The version I got is clearly marked IR-CUT and all of the descriptions/reviews/pictures suggest it's good at daylight without IR hue. It might be the same exact camera and only difference in lens (plus, 175 deg vs 160, as I may need the extra), but we'll see (the filter is there however, since it can be heard when switching the mode). I'll try it in a bit.


    > I tried yesterday again setting the exposure to manual with a fresh hyperbian install and now it is working fine.

    @lunchow3cu is that for Picam or C270? :huh:

  • > I'm using the Logitech C270 Cam
    Those are some good results, for sure!


    I got my PiCam working finally, and while the overall quality of the image is good (and I haven't seen "glare" or "banding" issues mentioned by the OP), looks like it's actually NOT a 175-degree fisheye :-/


    The zoom factor appears to make it closer somewhere to 120 degrees.. This is judging by the minimal distance I have to put it away from the TV so that the entire screen is in the FoV. Either that or I misunderstood how this works. In any case, I might just try to return it and get the other one mentioned above (160-degree, if it's actually true this time).

  • OK, finally found what I was looking for!

    Arducam Fisheye Camera Raspberry Pi - as one review says:

    Quote

    Many of the "fisheye" or "wide angle" raspberry camera + lens combos you'll find for sale tend to come up short in terms of field of view. The OV5647 sensor is a 1/4" image format sensor, and often there will be 1/3" format lenses mounted on them, which means the lens is focusing on a larger area than the sensor is capturing. So if it says it has 170 degree field of view lens, often the actual image will only be 140-150 degrees.


    That is not the case here. This actually has a field of view greater than 180 degrees; the plastic behind the camera actually shows up in the shot.

    and then it shows a picture that really does have circumference in the view. Ordered this instead and returning my current one.


    EDIT: Ordered this one as well and will compare them: Dorhea Raspberry Camera Fisheye Megapixel . Besides these 2, I haven't found any other ones that are truly at/over 180deg (like Ambivision).


    So I think with one of these two, I have a decent chance of doing a 730p@49fps video and straightening it with OpenCV fisheye module in real-time on Pi3B+. Everything points to this being sufficient in power.


    EDIT2: I think this might also be part of the reason why Govee wasn't developed for 75+ inch TVs. Their camera is wide-angle but not enough for such diagonal (exactly my case here). On top of that I suspect they really suffer from IR hue because their default color is pink (unless this is somehow just marketing or poor coding).

  • luchow3cu i need to install Journey to compare my result with yours. But on TV show dark brown a pretty well rendered.

    Do you have 30 or 60 led/meters in your config? Is it a LG CX TV :p


    With the last camera settings i published i think it is very accurate (only pastel colors seems not renderer, but i had some backlight, i don't want to watch TV in a too dark room.


    dinvlad I'm curious to see your results with openCV, it was my first idea before noticing i could place the camera far from TV.

  • David Hequet thanks! I think I got it from your earlier post, but then kinda forgot and found a tutorial where folks have done this for other use cases on RPi ^^ And I'm also on CX (77") so really looking forward to see the results too!

  • 77" !!! Ok you need maxi fisheye for this :p


    i was looking for 110°/120° fisheye with my liiiittlle 55CX


    Oled + Amiblight = wahou , what led density did you choose, i recycled my "old" ambilight 30 led/meters stripes and the brightness of the oled reduced the wow effect of the ambilight.. Thinking of buying 60 led/meters stripes

  • > Oled + Amiblight = wahou , what led density did you choose, i recycled my "old"

    David Hequet definitiely 60/m (at least) ^^ Planning to get ws2815 actually, because of its unique reliability (backup signal) feature. Another perk is ordering it in a Neon rope so that it looks exactly (or better!) like Hue Gradient 8)

    Something like this https://www.superlightingled.c…roll-for-sale-p-2929.html - can even mount it the same way by adopting one of these: Bewinner Bending Acrylic Suitable Cooling .


    But all of this after making the camera work first - everything else should be smooth sailing. In the end it will look like Hue, act like Ambivision and be more responsive and like 2x cheaper - let's see if any of that holds true ^^

  • Hmm, while waiting on 180+ cameras I've hit another snag - the surface of the TV is quite reflective, so it shows my floor lamp next to the TV at the edge of the screen, when I put the camera too close (~6-8 in. from the center). Not sure how to deal with this one - maybe "dark screen" frame subtraction would work here?

  • OK, got the fisheye Arducam and it's almost perfect for our needs! The FoV is a bit larger than my screen, but not overly periscope-large like the other one I received. I think this one can work for an even bigger TV! I can capture cropped video, which should bump up the framerate a little (~57 fps from a quick test).


    The only issue is the reflective surface as mentioned above, but that might be a minor nuisance. Can play around with background subtraction if needed, or leave it as-is in worst case. Or just turn off the lamp, since it should look the best in darkness anyways ^^

  • OK, got the fisheye Arducam and it's almost perfect for our needs! The FoV is a bit larger than my screen, but not overly periscope-large like the other one I received. I think this one can work for an even bigger TV! I can capture cropped video, which should bump up the framerate a little (~57 fps from a quick test).


    The only issue is the reflective surface as mentioned above, but that might be a minor nuisance. Can play around with background subtraction if needed, or leave it as-is in worst case. Or just turn off the lamp, since it should look the best in darkness anyways ^^

    I am happy so hear that. I just a china capture card for now, but might give the fisheye arducam a shot. How did you adjust the rectangular shape in the settings to the more curvy outline of the screen without manually placing each led/pixel?

  • Quote

    How did you adjust the rectangular shape in the settings to the more curvy outline of the screen without manually placing each led/pixel?

    That is what I'm currently working on! Using OpenCV fisheye module, something similar to https://medium.com/vacatronics…v-and-python-395528a51615


    The ultimate setup will be very simple:

    1) run a one-time calibration script, which will display a ChArUco pattern on the screen, and we just move the camera around at various angles pointing at the pattern, until the algorithm is "trained" sufficiently; this script will store calibration parameters into a JSON file;

    2) run a "live capture" script pointing to the JSON, which will continuously acquire the images, de-fisheye them so they look straight, and write the result to v4l2loopback interface;

    3) use the loopback interface as a regular camera source in Hyperion


    To make it more performant, I've just cross-compiled OpenCV for Pi with CPU optimizations (NEON etc). ?That'll be sufficient. From what I read, step 2 (which is the most crucial one for performance) doesn't require much processing power, since it's just a simple table lookup (and we're using optimized library). It will likely require Pi 3B+ though.. I have both that and Zero so we'll see.


    That being said, you can probably totally wing it and just use a trapezoid (I think HyperionNG already supports that) ^^ It'll be less precise, but might be "good enough".


    EDIT2: Thinking more about it, I'll probably also make it do crop and perspective-transform (if not too CPU-heavy) so there's no need to align corners in Hyperion. Should be pretty straighforward, and can be part of the calibration script.

Participate now!

Don’t have an account yet? Register yourself now and be a part of our community!