COLOR SIGNATURE AFFECT OTHERS

[East Yuan]

HI,

 every body , I found if I set 7 signatures ,then the color will affect each other . for example  , If I set dark yellow as sig1 , light yellow as sig2 ,then the pixy will not  recognize the light yellow one. other situation like green will affect red recognization . if I set only 1 signature ,then the pixy will recognize the light yellow one .so my question is can we change something , so the signature will not affect each other? 

Thanks 

Yuan

[Rich LeGrand]

Hello,
Yes, it is true that if you have color signatures that are close
together in hue-space, Pixy will give the signature with the lower
number priority. For example, if you taught Pixy the same object for
signature 1 and 2, signature 1 would dominate.

Significant overlap in hue space between signatures should be avoided.

thanks,
--rich

> --
> You received this message because you are subscribed to the Google Groups
> "pixy_firmware_sdk" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to pixy_firmware_...@googlegroups.com.
> To post to this group, send email to pixy_fir...@googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/pixy_firmware_sdk/44bdf3b4-7170-4684-a9d5-ce60310b2231%40googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.



--
Charmed Labs
www.charmedlabs.com

[East Yuan]

Rich:

Thanks for your reply.

I also test Hmulet program, if i use a pattern ,it works well .if I used a real pattern ,then it will not works as well.especially it will not recognize purple. I tested If there is only one signiture, then in most situation, it works well. so I think ,if it is possible in mo side ,each frame only check one signiture ? in m4 side finally deal with 7 result?

thanks 

Yuan

在 2016年11月13日星期日 UTC+8下午8:40:28，East Yuan写道：

[Rich LeGrand]

Hi Yuan,
Bear in mind that hue can vary quite a bit within an object and in
different lighting. Pixy's color signatures take this into account by
detecting a range of hues at different brightnesses If you're trying
to make distinctions between two objects that have similar hue, you
will probably find that it's difficult to do so robustly.

--rich

> --
> You received this message because you are subscribed to the Google Groups
> "pixy_firmware_sdk" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to pixy_firmware_...@googlegroups.com.
> To post to this group, send email to pixy_fir...@googlegroups.com.
> To view this discussion on the web visit

> https://groups.google.com/d/msgid/pixy_firmware_sdk/160025e4-ca51-4a68-b41a-fdac37a6c806%40googlegroups.com.

[East Yuan]

That  is corect! if change light ,then the result quite different . that means ov9715 dynamic range is a little bit lower , it is only 69 db. so how about we change the ov9715 to ov10635 ,then the dynamic range can improve to 115 db, the pixy can be High Dynamic Range .(HDR), and ov10635 can also direct output YUV signal . 

Best Regards

Yuan

在 2016年11月13日星期日 UTC+8下午8:40:28，East Yuan写道：

HI,

[Rich LeGrand]

Hi Yuan,
It's true that if you have 7 signatures loaded, they may affect each
other. This "cross-affecting" is worse when you teach Pixy objects
that have 2 or more "close" hues -- bluish-green and green, for
example. Or dark blue and dark purple. Pixy will prioritize the
lower-numbered signatures (allocate more space in the color space).
The result is that the higher-numbered signatures will have lower
detection accuracy. It can sometimes be a challenge to find 7 colors
that are distinct enough such that they don't interfere with each
other. This gets to the root of the task -- trying to detect objects
based on their hue.

Hope this helps.

--rich

On Mon, Dec 18, 2017 at 2:57 AM, East Yuan <yuany...@gmail.com> wrote:
> Hello every body.
> when I try to use pixy in a industry project ,then I found some problem.
> 1: the 9715 dynamic range is not enough ,it is only 69 db about , if it
> detect the white one good enough ,then there will be problem to detect dark
> object
> 2: because pixy is using on chip sram , it is only 260k, so the memory is
> not enough to do more complicate calculation.
> 3: the display size in pc side is only 320x200 , so it is too low, I expect
> 1280x800 .
>
> so I hope somebody help me to develop new hardware and software:
>
> my proposal:
> 1: using high dynamic range cmos chip like 10635.,the HDR camera.
> 2:can direct access YUV instead of RGB
> 3:with more sdram like 16MB
> 4:output display size can reach 1280x800

>
>
>
>
>
>
> 在 2016年11月13日星期日 UTC+8下午8:40:28，East Yuan写道：
>>

> --
> You received this message because you are subscribed to the Google Groups
> "pixy_firmware_sdk" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to pixy_firmware_...@googlegroups.com.
> To post to this group, send email to pixy_fir...@googlegroups.com.
> To view this discussion on the web visit

> https://groups.google.com/d/msgid/pixy_firmware_sdk/28a27aa7-012d-4184-991e-5cef310544d4%40googlegroups.com.