Node Red with OpenCV? Anyone played with it?

[Simon H]

Hi All,

I'm running a python motion detection program using OpenCV on my RPi3, but increasingly am finding myself hacking together interfaces to it so I can use Node-Red to control and display things (e.g. using bigtimer to stop it detecting motion during the day... using NR to serve up all the images, etc.).

Has anyone used OpenCV (e.g. interfaced via node-opencv) with NR directly?  I've searched and not found any obvious examples...

s

[Zenofmud]

did you try going to the node-red flows page and entering ‘opencv’?

--
http://nodered.org
 
Join us on Slack to continue the conversation: http://nodered.org/slack
---
You received this message because you are subscribed to the Google Groups "Node-RED" group.
To unsubscribe from this group and stop receiving emails from it, send an email to node-red+u...@googlegroups.com.
To post to this group, send email to node...@googlegroups.com.
Visit this group at https://groups.google.com/group/node-red.
To view this discussion on the web, visit https://groups.google.com/d/msgid/node-red/d126a305-8e16-43cc-95ed-991a58f3dd26%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Zenofmud]

Never mind, a funny a google search of node-red and opencv finds https://www.npmjs.com/package/node-red-contrib-opencv

but it is not showing on the flows page…

On Oct 15, 2017, at 12:19 PM, 'Simon H' via Node-RED <node...@googlegroups.com> wrote:

[Simon H]

"zuhito published 3 hours ago" - maybe my search was just a little early!!!

[Simon H]

For anyone following this; see Here for my log of trying to get opencv and node-red to work together on RPi3.  

Not got it quite to motion detection stage, but did manage to write a .jpg file derived from /dev/video0 from node-red, so I feel I'm slowly getting there.

'Getting there' did involve recovering my pi from a backup image after losing sudo - be careful of advice given about npm global installations.

[Bart Butenaers]

Hey Simon,

about your OpenCv question in the other discussion, I will answer it here to keep it all nice together.

Just like you I have been experimenting with the Javascript bindings for OpenCv, but I got mentally ill while trying to install OpenCv on my Raspberry Pi ;-)  

I think I have found a solution today today for this issue.  But it might become a complete failure...

Give me a couple of days to create a prove of concept, and I will get back here with my (hopefully positive) feedback.

Bart

[Simon H]

Hi Bart,
would be good to share results :)
Since I went for OpenCV 3 (not my choice; I just followed the install
procedure highlighted in my blog....), I would really like to hear
about OpenCV 2, for which 'normal' node-opencv is available; to know
if I should change. But node-opencv seems to still be in a big state
of flux... looks like some of the functions I want to use would need
to be added. Also, if you find a good node-opencv API description (or
a reliable way of dumping the API of an object), then let me know.
I've had a lot of complete failures with opencv :). Not the least
being an unstable pi... but it could be this is caused by writing lots
of images to a weak SD.
regards and luck :).
s

> --
> http://nodered.org
>
> Join us on Slack to continue the conversation: http://nodered.org/slack
> ---

> You received this message because you are subscribed to a topic in the
> Google Groups "Node-RED" group.
> To unsubscribe from this topic, visit
> https://groups.google.com/d/topic/node-red/GQeDS-CuK00/unsubscribe.
> To unsubscribe from this group and all its topics, send an email to

> node-red+u...@googlegroups.com.
> To post to this group, send email to node...@googlegroups.com.
> Visit this group at https://groups.google.com/group/node-red.
> To view this discussion on the web, visit

> https://groups.google.com/d/msgid/node-red/284d0598-8433-4d7e-9510-1c6968ccd2bf%40googlegroups.com.

[Bart Butenaers]

Hi Simon,

Let's share some 'intermediate' results, so at least you know in which direction I'm thinking.

As mentioned before, I would like to develop some OpenCv based Node-Red nodes (face detection, motion detection, ...).

However when the users would have to build and install the C++ libraries manually, the following would happen:

• 72% of the users would be to scared to download my nodes
• 23% of the users would get a depression after two evenings trying 
• 5% of the users would complete the installation and enjoy my new contribution

So I'm not interested in the C++ stuff currently.  Just want OpenCv to be installed automatically, as an NPM depedency of my node.

I have explained in this article how I want to accomplish that. 

It will be a little bit slower as the C++ binaries, so the 5% of my users would become disappointed.  But the other 95% would become excited ;-)

Please let me know what you about my idea!

I wanted to develop this contribution as prove of concept.  However did not have enough time to get it working yet.

It would be great if you could assist me in this development.  I have some issues with the object_detector.js file:

• At line 17 the 'ret' is always false.  OpenCv cannot find my xml file, even when I specify the full path

        var ret = faceClassifier.load('./classifiers/haarcascade_frontalface_default.xml');

• At line 68 an exception is raised (I assume because the xml file hasn't been loaded on line 17).  However I get this one:

Exception catching is disabled, this exception cannot be caught. Compile with -s DISABLE_EXCEPTION_CATCHING=0 or DISABLE_EXCEPTION_CATCHING=2 to catch.

 So I assume this need to be fixed in the opencv.js NPM package or I have to build Opencv myself ...

• I use jpeg-js for encoding and decoding jpeg (similar to the official example).  However I have to check whether OpenCv.js does support this kind of stuff also.
  
• Since OpenCv works with raw image data, I decode the input jpeg from the input message.  And afterwards I encode the raw image to jpeg for the output message.  However if we would chain a series of X Node-Red (OpenCv related nodes) in series, it would be more performant to do it like this: 

HttpRequestNode (to get jpeg from camera)

 --> DecoderNode (decode jpeg/tif/gif ... to raw image)

 --> SomeOpenCvNode (process the raw image)

 --> AnotherOpenCvNode (process the raw image)

 --> EncoderNode (encode raw image to jpeg, png...)

This way not every OpenCv node does encoding and decoding, but they pass simply rawimagedata over the Node-REd wires.  That would result in better performance (single encode and decode) and encoding/decoding logic does not need to be repeated in every OpenCv node. 

All help, thoughts and coding is appreciated !!

Bart

[Walter Kraembring]

Hi, I am using video motion monitoring with NR but not with OpenCV. Instead I use Motion, very powerful and easy to install, not necessary to build anything if you would like to run it in a RPi

https://htmlpreview.github.io/?https://github.com/Motion-Project/motion/blob/release-4.0.1/motion_guide.html#Motion_Overview

Installing with apt

Motion is part of the Ubuntu and Debian repositories and can be installed with the apt tool.
Open up a terminal window and type: sudo apt-get install motion
Before we start configuring Motion, we will copy the config file to our Home folder so that the master copy won't be affected. Open a terminal and copy the configuration file to your Home folder with following commands:
mkdir .motion
(Note: This will create a hidden folder .motion in your Home directory.)
cp /etc/motion/motion-dist.conf ~/.motion/motion.conf
(Note: This command will copy the original motion configuration file to its location.)
Now can open the configuration file for editing:
sudo nano ~/.motion/motion.conf
After you you have made all the appropriate adjustments to the configuration file, close it. Then start motion in the terminal simply by typing:
motion

Interfacing with NR: I use python scripts running in the Pi's interfacing between NR and Motion like this:

NR -> MQTT -> Python Scripts -> Http requests (URL's) for pausing or starting motion detection (can be used with BigTimer or other conditions decided by NR)

NR <- MQTT <- Python Scripts <- Executed by Motion on Events (motion detected, camera signal lost etc)

In addition, I have a cleanup script running once a day, deleting stored pictures and videos older than x days

[Bart Butenaers]

Hi Walter,

Indeed the Motion project will be a good solution for many users. No arguments against that...

But I would really want to use OpenCv for more advanced things.  For example face recognition, histogram calculations, ...

Probably Motion will be faster (since it is native C++) than OpenCv.js, but I still have to experiment with it.  

Perhaps the performance of OpenCv.js is really awfull, when running it on NodeJS (V8 engine). 

When reading articles, seems that the performance depends heavily on the context (FireFox, Chrome, NodeJs, ...).

Only time will tell ;-)

Thanks !

Bart

Virusvrij. www.avast.com

--

http://nodered.org
 
Join us on Slack to continue the conversation: http://nodered.org/slack
---
You received this message because you are subscribed to a topic in the Google Groups "Node-RED" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/node-red/GQeDS-CuK00/unsubscribe.

To unsubscribe from this group and all its topics, send an email to node-red+unsubscribe@googlegroups.com.

To post to this group, send email to node...@googlegroups.com.
Visit this group at https://groups.google.com/group/node-red.

To view this discussion on the web, visit https://groups.google.com/d/msgid/node-red/8c6bcd19-4221-43ed-842b-ce39bd62bf0b%40googlegroups.com.

[Simon H]

Hi Bart,

No, I didn't mean the motion project!....  I meant converting my current motion sense written in Python to NR using node-opencv....

Anyway; using Opencv 3 and this repo, I've now got basic motion sensing working.  The biggest issue I have is working what the API is...  so I cloned the repo and put it up again as a NEW push to github here - because forked repos are not searchable (???!!!).

The second biggest issue is if you do anything wrong, NR crashes.

The third issue to know about is you MUST release any images...  i.e. img.release(), else you run out of ram.  node-opencv does not (in a timely manner?) remove the video frame data on de-reference of a matrix object.

At full chat, the pi (model 3) manages ~8fps.  I put a 100ms delay in ->~5fps, but still very busy (75%); all on one core.

The test flow is here... I don't claim anything for the flow except that it works....

best regards,

Simon

To unsubscribe from this group and all its topics, send an email to node-red+u...@googlegroups.com.

To post to this group, send email to node...@googlegroups.com.
Visit this group at https://groups.google.com/group/node-red.

[Simon H]

ha-ha - sorry Walter!

@bart will review your post in detail later.  pm me; a call may be appropriate....

s

[Walter Kraembring]

ha-ha - sorry Walter!

??? I'm not sorry, in fact very happy,

I just saw you struggling, thought u would be interested in a smoother solution for video motion detection but if you find the tech part amusing or need some special features only provided by OpenCV, I understand

[Simon H]

I was sorry for being so preoccupied that I read Bart's response as being to me :). That's what trying to get opencv to work does to you!

In terms of the motion project; does it have a facility to mask the original image before detection motion?  I had imagined not, but always worth a look...

I'm trying to monitor a pond; and the ripples on the water are a little bit of a challenge - especially when the sun shines; 'sparkles' on the water cause continual false detection, so I mask the areas of the incoming image to exclude the water itself.

I'd also like (in the longer term) to specifically detect pigeons in the images tagged as motion, and exclude these as false positives (whilst not rejecting herons!) - I think this could be a tall order since I don't have many images of herons at the pond to train a detector with.... (I can get MANY pictures of pigeons, but since they are both birds, I suspect and pre-trained network would detect either just as well; and since I'm trying to find out what is taking my fish, I don't want to miss a heron, but am a bit sick of trawling through hundreds of pictures of pigeons). 

One clear advantage of using motion for the detection is that it would run on a different core, allowing NR more CPU time to do other things; I'd resisted the temptation, but now having got my fps results, may re-assess - I may still need opencv to analyse the images after motion has been detected, but it then becomes more of an offline process.

best regards,

Simon

[Julian Knight]

I suspect that you are asking a lot from a Pi, especially one running other services too. Personally, I'd want to run the camera input and motion detection on a separate device. I might also look at some non-Pi SBCs to get something with a bit more grunt. Peter Scargill's blog has compared some of those in the past.

[Colin Law]

On 23 October 2017 at 08:13, 'Simon H' via Node-RED

<node...@googlegroups.com> wrote:
> I was sorry for being so preoccupied that I read Bart's response as being to
> me :). That's what trying to get opencv to work does to you!
>
> In terms of the motion project; does it have a facility to mask the original
> image before detection motion? I had imagined not, but always worth a
> look...
> I'm trying to monitor a pond; and the ripples on the water are a little bit
> of a challenge - especially when the sun shines; 'sparkles' on the water
> cause continual false detection, so I mask the areas of the incoming image
> to exclude the water itself.

I use motion for very similar purposes. You can provide a mask file to
mask out areas on a pixel by pixel basis (you provide a black/white
image and it ignores movement in the black area).

> I'd also like (in the longer term) to specifically detect pigeons in the
> images tagged as motion, and exclude these as false positives (whilst not
> rejecting herons!) - I think this could be a tall order since I don't have
> many images of herons at the pond to train a detector with.... (I can get
> MANY pictures of pigeons, but since they are both birds, I suspect and
> pre-trained network would detect either just as well; and since I'm trying
> to find out what is taking my fish, I don't want to miss a heron, but am a
> bit sick of trawling through hundreds of pictures of pigeons).

Motion will not do the recognition for you.

>
> One clear advantage of using motion for the detection is that it would run
> on a different core, allowing NR more CPU time to do other things; I'd
> resisted the temptation, but now having got my fps results, may re-assess -
> I may still need opencv to analyse the images after motion has been
> detected, but it then becomes more of an offline process.

Whether the pi will do it depends on the image size, picture rate, and
whether it has to decompress the image (if it is coming from an rtsp
stream for example). I know I couldn't get away with a pi, but then I
am monitoring a 1024x768 rtsp stream running at 25 fps and checking
for movement at 12fps. If it were a 640x480 mpjpeg stream checking for
movement at 2fps then it might do it.

Colin

[Bart Butenaers]

Hi folks,

indeed my Raspberry won't find this kind of computing amusing.  But like Walter described : I indeed really find the tech part amusing...

So I will continue my search for creating OpenCv.js based nodes, to make sure that even my grandfather can add heavy image processing to his Node-Red flows :-)

I have added a Canny Edge detector test to my experimental contribution. 

So now I can call OpenCv from Node-Red without having to install it.  Woohoo.  

But it is much slower than the Asm.js benchmarks predict.  Damn.

You can see on my page that I have created an issue for the OpenCv.js team, but don't expect an answer soon (when looking at their other issues).

However due to unexpected medical reasons, I will have to drop most of my Node-Red stuff for some time.

Will continue my experiment later on.  Will keep you guys informed here when I'm back in town ...

Keep posting good information !!

Bart

[Dave C-J]

While def not node.js, this guy is THE person to read up on opencv and its capabilities.. Easy to read and follow. And includes pi as well https://www.pyimagesearch.com/2017/09/04/raspbian-stretch-install-opencv-3-python-on-your-raspberry-pi/

[Simon H]

:). it's his python code I ported.... highly recommended site,
although if you sign up you do get quite a few emails (all fairly
relevant).

> --
> http://nodered.org
>
> Join us on Slack to continue the conversation: http://nodered.org/slack
> ---
> You received this message because you are subscribed to a topic in the
> Google Groups "Node-RED" group.
> To unsubscribe from this topic, visit
> https://groups.google.com/d/topic/node-red/GQeDS-CuK00/unsubscribe.
> To unsubscribe from this group and all its topics, send an email to
> node-red+u...@googlegroups.com.
> To post to this group, send email to node...@googlegroups.com.
> Visit this group at https://groups.google.com/group/node-red.
> To view this discussion on the web, visit

> https://groups.google.com/d/msgid/node-red/CACXWFwLxHdxsJXbC6tkHUgT%2BNXeM%3DHDSZzL8FRxY6ov%2B5wHCFw%40mail.gmail.com.

[Simon H]

ok, playing some more; implemented bart's multi-part-encoder for extracting the images as a mjpeg feed...

I'me now doing motion detection at the full 30fps by reducing the image size down to 300 width (4x3 aspect).

But when the jpeg encoding is enabled, it drops to ~20fps (encoding a 640x480 and sending via the multipart encoder)...

The pi CPU is reporting 25% (i.e. one core fully occupied...)

So I'll start a separate thread about using two NR instances together with raw data going between them...

s

[Julian Knight]

Are you sure you want the raw data going between the Pi's? Perhaps with some thought you could keep the image data and processing on 1 Pi and just use the other for control using MQTT & Node-RED?

[Simon H]

hi julian,

was thinking more of two instance on the one pi; memory permitting (it's a 3, so 1gb).... just so I can start using a 2nd cpu core.

s

[Julian Knight]

In that case, a UNIX Pipe would be the most efficient. But I think you will be pushing that 1GB of RAM.

[Simon H]

memory is of concern; 1 x NR = ~256 mbytes ram, according to my current estimations.

[Zenofmud]

Just tossing out an idea...could you do this running a pi cluster??? (Hmmm could node-red even run on a cluster....)

Paul

--

http://nodered.org
 
Join us on Slack to continue the conversation: http://nodered.org/slack
---

You received this message because you are subscribed to the Google Groups "Node-RED" group.

To unsubscribe from this group and stop receiving emails from it, send an email to node-red+u...@googlegroups.com.

To post to this group, send email to node...@googlegroups.com.
Visit this group at https://groups.google.com/group/node-red.

To view this discussion on the web, visit https://groups.google.com/d/msgid/node-red/3734d0a6-e03f-42ce-8501-09d91f7812a7%40googlegroups.com.

[Colin Law]

I don't think a cluster would help with node red itself as it is
single threaded. Intensive processes spawned by node red could be
handled by a cluster however.

Colin

> https://groups.google.com/d/msgid/node-red/617E50F5-4226-4E40-8D4B-BD23F1F59BDA%40zenofmud.org.

[Simon H]

Sorry for posting in multiple places, but for the node-opencv record:

There are some functions in the node-opencv that I'm using which use Nan::AsyncQueueWorker to launch a separate thread at the C++ level.

Testing indicates that this does indeed allow the CPU to exceed 1 core's capacity (have seen NR using 220% of the 400% available on the RPi3).

By searching my copy of the github for AsyncQueueWorker, I find the following functions should run a separate thread:

cv.readImage

videocapture.read

videocapture.grab

videocapture.retrieve

matrix.saveAsync

matrix.toBufferAsync

features.similarity

CascadeClassifierWrap::DetectMultiScale

FaceRecognizerWrap::Train

FaceRecognizerWrap::Predict

I don't think the official node-opencv has more....

It is a pity more matrix operations are not async (e.g. image scaling).

Given the time, I could adapt the node-opencv to add more async functions, after analysing my flows for where most benefit would be had...

s

[Simon H]

For anyone following use of OpenCV in Node-Red, I've started some pages on it here:

btsimonh.wordpress.com/node-opencv-with-node-red/

Mainly focused around RPi3, opencv 3.3.0, node-opencv, motion detection, and of course, Node-Red.

[Simon H]

published an example flow using backgroundsubtractor for motion detection in NR

https://flows.nodered.org/flow/33a93ac5418009993d38c00009ef453e

[wku...@gmail.com]

Liked your blog, you might want to add this to the instructions to "automagically" start bcm2835-v4l2 module:

sudo nano /etc/modules

// add as the last line:

bcm2835-v4l2

// save and exit nano

[Simon H]

thankyou!  updated the page in the blog - will try this tonight on my own pi....

[Simon H]

Next installment - motion detection with video recording of motion detected!

https://btsimonh.wordpress.com/video-recording-from-node-red-using-motion-detection-and-node-opencv/

[Bart Butenaers]

Simon,

Have found it.  Forgot to add the package name 'opencv' to the command:

npm install opencv --nodeDir=/home/pi/node-opencv-nr/node-opencv-nr

How stupid of me.  But now Node-Red starts without errors, and a /home/pi/.node-red/node_modules/opencv/ folder is created, as expected.

Some remarks about your first (installation) blog:

• perhaps it is useful to add the above command to your blog, for users that want to install it locallly ?
• In the command sudo chown pi:pi /usr/lib/node-modules there was an issue for me.  In my case it is node_modules, instead of node-modules ( _ instead of - ).  Is this a typo ?
• Perhaps you can add some (optional) commands to remove the zip files and src folders.  Not really required, but it could keep your RPI clean ...
• The part of your own fork was in the beginning not clear to me.  Perhaps you can just say that the need to use git clone https://github.com/btsimonh/node-opencv-nr if they want your latest developments, and that they need to do git clone https://github.com/peterbraden/node-opencv if they want the official version.
• For the settings file it might be usefull to explain what they have to add:

Keep up the good work !

Bart

On Sat, Nov 4, 2017 at 8:59 PM, 'Simon H' via Node-RED <node...@googlegroups.com> wrote:

Next installment - motion detection with video recording of motion detected!

https://btsimonh.wordpress.com/video-recording-from-node-red-using-motion-detection-and-node-opencv/

--
http://nodered.org
 
Join us on Slack to continue the conversation: http://nodered.org/slack
---

You received this message because you are subscribed to a topic in the Google Groups "Node-RED" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/node-red/GQeDS-CuK00/unsubscribe.
To unsubscribe from this group and all its topics, send an email to node-red+unsubscribe@googlegroups.com.

To post to this group, send email to node...@googlegroups.com.
Visit this group at https://groups.google.com/group/node-red.

To view this discussion on the web, visit https://groups.google.com/d/msgid/node-red/dedbf6dd-de3d-4fe1-abe6-7a4548704c6b%40googlegroups.com.

[Bart Butenaers]

Simon,

Sorry to disturb you, but I think you might have to pimp the function node coding from your first blog.

It contains some syntax errors:

Bart

On Sat, Nov 4, 2017 at 8:59 PM, 'Simon H' via Node-RED <node...@googlegroups.com> wrote:

Next installment - motion detection with video recording of motion detected!

https://btsimonh.wordpress.com/video-recording-from-node-red-using-motion-detection-and-node-opencv/

--
http://nodered.org
 
Join us on Slack to continue the conversation: http://nodered.org/slack
---

You received this message because you are subscribed to a topic in the Google Groups "Node-RED" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/node-red/GQeDS-CuK00/unsubscribe.
To unsubscribe from this group and all its topics, send an email to node-red+unsubscribe@googlegroups.com.

To post to this group, send email to node...@googlegroups.com.
Visit this group at https://groups.google.com/group/node-red.

To view this discussion on the web, visit https://groups.google.com/d/msgid/node-red/dedbf6dd-de3d-4fe1-abe6-7a4548704c6b%40googlegroups.com.

[Simon H]

@bart

thanks for the feedback; keep it coming :)

fixed node-modules to node_modules.

Updated the bit about settings.js to be more explicit about require (i.e. providing my functionGlobalContext section).

Added a 'Quick install' section, which skips all the git stuff, and just installs directly from my github branch :).

Added option for local installation in .node-red folder.

Updated both the flows in the flow library to allow global or locally installed node-opencv, and error if require is not found.

hmm... just copied the first flow from the page and imported in NR, and all is ok...

Certainly with the longer flows the WP site is a right pain in the a**e, which is why I put the later flows in NR flows library.

Itching to start on a node-red-contrib-xxx - but instead trying to talk to zuhito about collaborating first:

(https://github.com/zuhito/node-red-contrib-opencv/issues/1)

s

[Zenofmud]

Do you save the flows surrounding them with <code>…</code> in WP?

[Simon H]

now all moved to gists :).
WP problems included:
truncation
edit view (or even reading view if you are logged in) not the same as other see.
quote replacement " and ' !!
corruption.

> --
> http://nodered.org
>
> Join us on Slack to continue the conversation: http://nodered.org/slack
> ---
> You received this message because you are subscribed to a topic in the
> Google Groups "Node-RED" group.
> To unsubscribe from this topic, visit
> https://groups.google.com/d/topic/node-red/GQeDS-CuK00/unsubscribe.
> To unsubscribe from this group and all its topics, send an email to

> node-red+u...@googlegroups.com.

> To post to this group, send email to node...@googlegroups.com.
> Visit this group at https://groups.google.com/group/node-red.
> To view this discussion on the web, visit

> https://groups.google.com/d/msgid/node-red/A3434E5C-B9E7-4992-8989-8BA53946C78F%40zenofmud.org.

[Zenofmud]

When creating or editing posts in WP, I always use ‘text’ instead of ‘visual’ and wrap code (like flows) in the <code>…</code> and never have an issue with truncation or the quotes and double quotes getting ‘smart’ (grin)

 

You received this message because you are subscribed to the Google Groups "Node-RED" group.
To unsubscribe from this group and stop receiving emails from it, send an email to node-red+u...@googlegroups.com.
To post to this group, send an email to node...@googlegroups.com.

Visit this group at https://groups.google.com/group/node-red.

To view this discussion on the web, visit https://groups.google.com/d/msgid/node-red/CAHt4oVWo-vT8v7cqe-16U29OTCteDFEtEPs9gi6ZA8%3D8B8-VYg%40mail.gmail.com.

[moorthy s]

Hi all,

Is OpenCV node available for Windows? 

This https://www.npmjs.com/package/node-red-contrib-opencv suggest it is only available for Rpi and Ubuntu.

I tried to install on my Win7 PC both via palette manager & cmd npm install method but wasn't able to..

My requirement is to find defective glasses from normal ones & i heard it is possible by training a classifier in [Tensorflow,Watson..etc]But how to do it with OpenCV node red?Any Suggestions?

[Dave C-J]

You wouldn't train the classifier in Node-RED - you would need to do that first. It may then be possible to run the model using OpenCV, which in turn can run Keras and Tensorflow models via Node-RED - but in general I would get them running standalone first - and then integrate them into Node-RED once you had the model running well.

[Julian Knight]

If you really need to run under Windows, try the Ubuntu subsystem for Windows (assuming you are using a version of Windows 10 from this year or later). You can install node.js and node-red in that and it works pretty much at native speeds.

[Simon H]

Yes, you can certainly run opencv and node-opencv in windows.
I moved my opencv nr dev to windows to simplify dev.
You can install the windows pre-build opencv, then set some env options to allow node-opencv to build. I've used 3.0 and 3.3.
The pre-built does not include the extras, but has basic functions.
Not got any experience with classifiers (yet)... but looking for described use cases for node-red-contrib-ocv (still in the design stage).