OpenCV with Kivy

[Richey]

Has anyone done any serious computer vision applications with Kivy?  I've got a project that I've built in python that uses OpenCV pretty heavily and would like to port it to Android and iOS.  Here is a trivial example:

#!/usr/bin/python

import cv2

import random

from numpy import array

cap = cv2.VideoCapture(0)

while True:

    initial_thresh = random.randint(100,170)

    ret,im = cap.read()

    im_gray = cv2.cvtColor(im,cv2.COLOR_BGR2GRAY)

    key = cv2.waitKey(10)&255

    cv2.imshow('gray',im_gray)

So, I looked at a Kivy example of getting the video stream found here: http://karanbalkar.com/2012/10/tutorial-6-working-with-camera-in-kivy/  This example doesn't show how to use cv2 to process the video stream through. Is this possible in Kivy? Will I be able to port a heavy OpenCV application to Android and iOS using Kivy?

Thanks,

Michael

[mathieu....@gmail.com]

Hi,

I'll be intereted in doing kivy + opencv too ! About your question, I've looked the source code of the camera_opencv.py in the kivy / core / camera folder.
There is the interaction with opencv ( the first python binding ). I don't think it'll be difficult to port it with the cv2 module.

Cheers !!

[Mathieu Virbel]

Hi,

I've done heavy computation using OpenCV within Kivy. No issues, it just works.

But never tried on mobile. Nobody wrote a recipe for android/ios yet, it need a hero here :)

--
You received this message because you are subscribed to the Google Groups "Kivy users support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kivy-users+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

[Jonathan Berte]

Hi,

Did people write a recipe for opencv on android in the meanwhile?

Bye,

jonathan

Op dinsdag 9 juli 2013 15:49:46 UTC+2 schreef Richey:

[Andreas Bunkahle]

I have a piece of code which gets an image from the OpenCV camera and puts it into a kivy texture to display. This means you have the possibility to do all kind of OpenCV transformations on a picture and put it to a kivy output later. The code looks like this:

__author__ = 'bunkus'
from kivy.app import App
from kivy.uix.widget import Widget
from kivy.uix.boxlayout import BoxLayout
from kivy.uix.image import Image
from kivy.clock import Clock
from kivy.graphics.texture import Texture

import cv2
import numpy as np

class CamApp(App):

    def build(self):
        self.img1 = Image(source='images/1.jpg')
        layout = BoxLayout()
        layout.add_widget(self.img1)
        #opencv2 stuffs
        self.capture = cv2.VideoCapture(0)
        ret, frame = self.capture.read()
        cv2.namedWindow("CV2 Image")
        cv2.imshow("CV2 Image", frame)
        Clock.schedule_interval(self.update, 1.0/33.0)
        return layout

    def CreateImage(self, (height, width), bits=np.uint8, channels=3, color=(0, 0, 0)): # (cv.GetSize(frame), 8, 3)
        """Create new image(numpy array) filled with certain color in RGB"""
        # Create black blank image
        if bits == 8:
            bits = np.uint8
        elif bits == 32:
            bits = np.float32
        elif bits == 64:
            bits = np.float64
        image = np.zeros((height, width, channels), bits)
        if color != (0, 0, 0):
            # Fill image with color
            image[:] = color
        return image

    def update(self, dt):
        # display image from cam in opencv window
        ret, frame = self.capture.read()
        cv2.imshow("CV2 Image", frame)
        # convert it to texture
        buf1 = cv2.flip(frame, 0)
        buf = buf1.tostring()
        texture1 = Texture.create(size=(frame.shape[1], frame.shape[0]), colorfmt='bgr')
        texture1.blit_buffer(buf, colorfmt='bgr', bufferfmt='ubyte')
        # display image from the texture
        self.img1.texture = texture1

if __name__ == '__main__':
    CamApp().run()

My question is:
What is the way vice versa, to get a camera picture from the kivy Cam and transform it to opencv? Has anyone code for this?

[Andreas Bunkahle]

Hi, me again. Finally I found a way to get the buffer of the texture to translate into a numpy array but I still get errors in the display (it is threefold). So obviously there is an error in the numpy transformation. Can anyone help?

from kivy.app import App
from kivy.uix.boxlayout import BoxLayout
from kivy.uix.button import Button
from kivy.uix.camera import Camera
from kivy.uix.widget import Widget
from kivy.clock import Clock
import cv2, numpy as np

class Cam(Camera):
    def build(self):
        pass
   
    def get_texture(self):
        # print "Texture", self.texture
        return self.texture
   
class CameraApp(App):
    def build(self):
        s = 100
        size = (s, s)
        sh = (None, None)
        self.snapshot = None
        root = Widget()
        self.width = 640
        self.height = 480
        self.channels = 3
        self.camera = Cam(resolution=(self.width, self.height), size=(self.width, self.height), pos=(100,100), play=True)
        root.add_widget(self.camera)
        buttons = BoxLayout(orientation='horizontal')
        sn = Button(text='snapshot', pos=(self.width/2, 20), halign='center', size_hint=sh, size=size)
        sn.bind(on_press=self.get_texture1)
        buttons.add_widget(sn)
        root.add_widget(buttons)
        # print "Texture", self.camera.get_texture, type(self.camera.get_texture), dir(self.camera.get_texture)
        Clock.schedule_interval(self.get_texture1, 1.0/33.0)
        return root

    def get_texture1(self, event):
        self.snapshot = self.camera.get_texture()
        # print "self.snapshot", self.snapshot, type(self.snapshot), dir(self.snapshot)
        self.reg = self.snapshot.get_region(0, 0, 50, 50)
        # print "self.reg.pixels", type(self.reg.pixels)
        # print self.reg, type(self.reg), dir(self.reg)
        frame = np.fromstring(self.snapshot.pixels,
                dtype=np.uint8,
                count=self.width*self.height*self.channels)
        frame = frame.reshape(self.height, self.width, self.channels)
        # print dir(frame)
        # print type(frame)
        cv2.imshow("Cv2", cv2.flip(frame, 0))
    
if __name__ == '__main__':
    CameraApp().run()

[llo...@gmail.com]

Easy stuff for opencv proffies

first you have to change the kivi texture to a string,
then you must create a numpy array from this string,
then you must reshape the array to a (R,G,B, Alpha) array

just check out this piece of code :

        nparr = np.fromstring(self.cam.texture.pixels, dtype=np.uint8)
        a = np.reshape(nparr, (480,640,4))
        #a = cv2.cvtColor(a, cv2.cv.CV_RGB2GRAY)
        cv2.imshow("test", a)


Greetings from Poland :)

[Danylo Ulianych]

@Andreas Bunkahle,

Thank you for sharing how to display OpenCV VideoCapture on Kivy! I'd only add

def close_settings(self, *largs):
    Clock.unschedule(self.update)
    del self.capture
    App.close_settings(self, *largs)

to override close_setting() to release properly VideoCapture after hitting an escape button (without managing so it continues running with a message "Leaving application in progress...")

Do you know how to set Window size initially in order to fit in with VideoCapture size (there is a black margin between capture frame and a window rectangle during displaying)? I found only one way to do such a thing: add

h = int(self.capture.get(cv2.CAP_PROP_FRAME_HEIGHT))
w = int(self.capture.get(cv2.CAP_PROP_FRAME_WIDTH))
Config.set('graphics', 'width', str(w))
Config.set('graphics', 'height', str(h))

before calling super().__init__(**kwards) in your CamApp.__init__(**kwards). But that's a bit nasty and I think there should be more elegant way to set a window size without changing default system window size.

[rhavel salviano]

Hello guys, I'm studying Kivy and made some changes to a code posted here. I wonder why the modified code does not display the camera?

from kivy.app import App, Widget

from kivy.uix.boxlayout import BoxLayout

from kivy.uix.image import Image

from kivy.clock import Clock

from kivy.graphics.texture import Texture

 

import cv2

import numpy as np

class test(Widget):

      

    def teste(self):      

        self.img1 = Image(source='images/1.jpg')

        layout =Widget()

        layout.add_widget(self.img1)

        #opencv2 stuffs

        self.capture = cv2.VideoCapture(0)

        ret, frame = self.capture.read()

        #cv2.namedWindow("CV2 Image")

        #cv2.imshow("CV2 Image", frame)

        Clock.schedule_interval(self.update, 1.0/33.0)

        return layout

    def update(self, dt):

        # display image from cam in opencv window

        ret, frame = self.capture.read()

        cv2.imshow("CV2 Image", frame)

        # convert it to texture

        buf1 = cv2.flip(frame, 0)

        buf = buf1.tostring()

        texture1 = Texture.create(size=(frame.shape[1], frame.shape[0]), colorfmt='bgr')

        texture1.blit_buffer(buf, colorfmt='bgr', bufferfmt='ubyte')

        # display image from the texture

        self.img1.texture = texture1

class CamApp(App):

    def build(self):

        return test()

[arnav goel]

Hello I have a kivy file and a python script. I would like to put the opencv video in the kivy file. Is there any way to do so . Also I tried using the camera widget but it says the resolution not supported. Whole problem on http://stackoverflow.com/questions/38446292/kivy-error-camera-webcam-gives-error-videocaptureresolution-not-found

If there is any way I could incorporate opencv video in my kivy application, It would be great. Thanks in advance. 

[Eugene Bartosh]

They support OpenCV 2.x only, no OpenCV 3.x support - 2.x is too old, that hardly makes sense (e.g. I use CLAHE in most of my projects which is missing in 2.x)

[Mayank Dhiman]

hey andreas... were you able to solve this problem?