Index>Robot Kit>RaspTank-Pro Smart Robot Car Kit for RPi>Lesson 12 Performing Color Detection with OpenCV

Lesson 12 Performing Color Detection with OpenCV


Lesson 11 Transmit Image in Real Time


This tutorial introduces how to transmit images in real time via the Raspberry Pi.

11.1 Components & Parts





Raspberry Pi



Robot HAT






Camera Flex Cable (black)




11.2 Transmitting via Flask-Video-Streaming


The Raspberry Pi robot features the real-time video and OpenCV functions. There are many methods of transmitting videos captured by the Raspberry Pi camera via network in a real time manner, and this tutorial introduces an open source project following the MIT License on Github:

The project uses Flask and related dependencies which have been included in the installation scripts for the Adeept robot. You may need to install them if your Raspberry Pi has run the script before.  

sudo pip3 install flask


sudo pip3 install flask_cors

The OpenCV part will not be involved here; the tutorial only introduces how to view the image of the Raspberry Pi camera on other devices in real time. First, download the flask-video-streaming project. You can clone on Github or download on your computer and transfer to the Raspberry Pi, using the command on Raspberry Pi Command Line:

sudo git clone

When the Raspberry Pi is configured with the robot software, the Raspberry Pi will automatically run the program. If you need to use the camera in other programs, you need to terminate this program. Termination command:

sudo killall python3

After flask-video-streaming is downloaded on the Raspberry Pi or transferred, run the file in the project:

cd flask-video-streaming


sudo python3

Pay attention not to run by the command "sudo python3 flask-video-streaming/", or there will be an error of unfound *.jpeg file.

Open a web browser (Chrome for example) on a device on the same LAN of the Raspberry Pi, enter in the address bar the Raspberry Pi's IP address and the video stream port number ":5000", as shown below:

Then you can view the webpage created by the Raspberry Pi on your mobile or computer. Note that by default, images of 3 numbers 1, 2, and 3 will loop instead of anything from the Raspberry Pi.


If you can log into the page and 1, 2, and 3 images loop display, it indicates the flask program runs well. Then you can change the file to display videos collected by the Raspberry Pi's camera.

Here we use the nano built in Raspbian to open and edit the There's no need to edit in other IDEs as only commenting or uncommenting involved.

sudo nano

1. Uncomment the code after opening

1. if os.environ.get('CAMERA'):  

2.     Camera = import_module('camera_' + os.environ['CAMERA']).Camera 

3. else:  

4.     from camera import Camera 

2. Add "#" at the beginning of the lines, or insert "'''" at the beginning and end of the paragraph to comment. The code is changed as follows:

1. # if os.environ.get('CAMERA'):  

2. #     Camera = import_module('camera_' + os.environ['CAMERA']).Camera  

3. # else:  

4. #     from camera import Camera  


1. '''  

2. f os.environ.get('CAMERA'):  

3.    Camera = import_module('camera_' + os.environ['CAMERA']).Camera  

4. lse:  

5.    from camera import Camera  

6. '''  

3. At last, uncomment the code Camera imported from camera_pi by deleting "#" – pay attention to delete the space after "#".

Code before change:

1. # from camera_pi import Camera  

Code changed:

1. from camera_pi import Camera  

4. Complete code of changed as follows:

1. #!/usr/bin/env python  

2. from importlib import import_module  

3. import os  

4. from flask import Flask, render_template, Response  


6. # import camera driver  

7. ''''' 

8. if os.environ.get('CAMERA'): 

9.     Camera = import_module('camera_' + os.environ['CAMERA']).Camera 

10. else: 

11.     from camera import Camera 

12. '''  


14. # Raspberry Pi camera module (requires picamera package)  

15. from camera_pi import Camera  


17. app = Flask(__name__)  



20. @app.route('/')  

21. def index():  

22.     """Video streaming home page."""  

23.     return render_template('index.html')  



26. def gen(camera):  

27.     """Video streaming generator function."""  

28.     while True:  

29.         frame = camera.get_frame()  

30.         yield (b'--frame\r\n'  

31.                b'Content-Type: image/jpeg\r\n\r\n' + frame + b'\r\n')  



34. @app.route('/video_feed')  

35. def video_feed():  

36.     """Video streaming route. Put this in the src attribute of an img tag."""  

37.     return Response(gen(Camera()),  

38.                     mimetype='multipart/x-mixed-replace; boundary=frame')  



41. if __name__ == '__main__':  

42.'', threaded=True)  

5. Press CTRL+X to exit after editing. A prompt will be shown asking you whether to save to not. Type in Y and press Enter to save.

6. Next, run

sudo python3

7. Open a web browser (here we use Chrome as an example) on a device on the same LAN of the Raspberry Pi, enter in the address bar the Raspberry Pi's IP address and the video stream port number ":5000", as shown below:

8. Now you can view the webpage created by the Raspberry Pi on your mobile or computer. After data is loaded successfully, it'll display the videos captured by the Raspberry Pi in real time.


9. This function is based on the flask-video-streaming project from GitHub:


11.3Processing Video Frames

Principle of Multithreaded Video Frames Processing

The OpenCV function is based on the flask-video-streaming project on GitHub; here we just changed the file for operations with OpenCV.

Single threaded video frames processing

Here we start with single threading for you to better understand why multithreading is needed for processing OpenCV video frames. The process for single threading is as follows:



Process explanation: First, capture an image frame from the camera, analyze the frame with OpenCV, generate the information to be drawn, like the central position of the target or the text to be displayed on the screen, draw accordingly, and then display the image which has been processed and drawn on the webpage.

This whole process is inefficient as it needs to wait the OpenCV to implement the processing and display on the screen for each frame before starting the next frame processing. It may cause a stuck video transmission.

Multi-threaded video frames processing

The process is as shown below:


Process explanation: To increase frame rate, here we separate the analyzing of video frames from the collection-display process and run on background thread to generate image drawing information.

The code is changed as below: (the OpenCV function is not included here to explain the multi-threaded processing principle; refer to the file in the zip file downloaded)

1. import os  

2. import cv2  

3. from base_camera import BaseCamera  

4. import numpy as np  

5. import datetime  

6. import time  

7. import threading  

8. import imutils  


10. class CVThread(threading.Thread):  

11.     ''''' 

12.     This class is used to process OpenCV's task of analyzing video frames in the background

13.     '''  

14.     def __init__(self, *args, **kwargs):  

15.         self.CVThreading = 0  


17.         super(CVThread, self).__init__(*args, **kwargs)  

18.         self.__flag = threading.Event()  

19.         self.__flag.clear()  



22.     def mode(self, imgInput):  

23.         ''''' 

24.         This method is used to pass in video frames that need to be processed

25.         '''  

26.         self.imgCV = imgInput  

27.         self.resume()  



30.     def elementDraw(self,imgInput):  

31.         ''''' 

32.         Draw elements on the screen

33.         '''  

34.         return imgInput  


36.     def doOpenCV(self, frame_image):  

37.         ''''' 

38.         Add content to be processed by OpenCV here 

39.         '''  

40.         self.pause()  



43.     def pause(self):  

44.         ''''' 

45.         Block the thread and wait for the next frame to be processed

46.         '''  

47.         self.__flag.clear()  

48.         self.CVThreading = 0  


50.     def resume(self):  

51.         ''''' 

52.         Resuming the thread

53.         '''  

54.         self.__flag.set()  


56.     def run(self):  

57.         ''''' 

58.         Processing video frames in a background thread

59.         '''  

60.         while 1:  

61.             self.__flag.wait()  

62.             self.CVThreading = 1  

63.             self.doOpenCV(self.imgCV)  



66. class Camera(BaseCamera):  

67.     video_source = 0  


69.     def __init__(self):  

70.         if os.environ.get('OPENCV_CAMERA_SOURCE'):  

71.             Camera.set_video_source(int(os.environ['OPENCV_CAMERA_SOURCE']))  

72.         super(Camera, self).__init__()  


74.     @staticmethod  

75.     def set_video_source(source):  

76.         Camera.video_source = source  


78.     @staticmethod  

79.     def frames():  

80.         camera = cv2.VideoCapture(Camera.video_source)  

81.         if not camera.isOpened():  

82.             raise RuntimeError('Could not start camera.')  

83.         ''''' 

84.         Instantiate CVThread()

85.         '''  

86.         cvt = CVThread()  

87.         cvt.start()  


89.         while True:  

90.             # read current frame  

91.             _, img =  


93.             if cvt.CVThreading:  

94.                 ''''' 

95.                If OpenCV is processing video frames, skip

96.                 '''  

97.                 pass  

98.             else:  

99.                 ''''' 

100.  If OpenCV is not processing video frames, give the thread that processes the video frame a new video frame and resume the processing thread

101.                 '''  

102.                 cvt.mode(img)  

103.                 cvt.resume()  

104.             ''''' 

105.             Draw elements on the screen 

106.             '''  

107.             img = cvt.elementDraw(img)  


109.             # encode as a jpeg image and return it  

110.             yield cv2.imencode('.jpg', img)[1].tobytes()  

That's the code for multi-threaded OpenCV processing. In the subsequent part of introducing details of the OpenCV function, we will only explain the method of video frame processing with OpenCV and skip this part.  

11.4 OpenCV Function

– First, create two .py files in a same folder of the Raspberry Pi (they are already included in the product download package for the Adeept Robot; refer to and, with code as shown below:

1. #!/usr/bin/env python3  


3. from importlib import import_module  

4. import os  

5. from flask import Flask, render_template, Response  


7. from camera_opencv import Camera  


9. app = Flask(__name__)  


11. def gen(camera):  

12.     while True:  

13.         frame = camera.get_frame()  

14.         yield (b'--frame\r\n'  

15.                b'Content-Type: image/jpeg\r\n\r\n' + frame + b'\r\n')  


17. @app.route('/')  

18. def video_feed():  

19.     return Response(gen(Camera()),  

20.                     mimetype='multipart/x-mixed-replace; boundary=frame')  



23. if __name__ == '__main__':  

24.'', threaded=True)

1. import time  

2. import threading  

3. try:  

4.     from greenlet import getcurrent as get_ident  

5. except ImportError:  

6.     try:  

7.         from thread import get_ident  

8.     except ImportError:  

9.         from _thread import get_ident  



12. class CameraEvent(object):  

13.     """An Event-like class that signals all active clients when a new frame is 

14.     available. 

15.     """  

16.     def __init__(self):  

17. = {}  


19.     def wait(self):  

20.         """Invoked from each client's thread to wait for the next frame."""  

21.         ident = get_ident()  

22.         if ident not in  

23.             # this is a new client  

24.             # add an entry for it in the dict  

25.             # each entry has two elements, a threading.Event() and a timestamp  

26.   [ident] = [threading.Event(), time.time()]  

27.         return[ident][0].wait()  


29.     def set(self):  

30.         """Invoked by the camera thread when a new frame is available."""  

31.         now = time.time()  

32.         remove = None  

33.         for ident, event in  

34.             if not event[0].isSet():  

35.                 # if this client's event is not set, then set it  

36.                 # also update the last set timestamp to now  

37.                 event[0].set()  

38.                 event[1] = now  

39.             else:  

40.                 # if the client's event is already set, it means the client  

41.                 # did not process a previous frame  

42.                 # if the event stays set for more than 5 seconds, then assume  

43.                 # the client is gone and remove it  

44.                 if now - event[1] > 5:  

45.                     remove = ident  

46.         if remove:  

47.             del[remove]  


49.     def clear(self):  

50.         """Invoked from each client's thread after a frame was processed."""  




54. class BaseCamera(object):  

55.     thread = None  # background thread that reads frames from camera  

56.     frame = None  # current frame is stored here by background thread  

57.     last_access = 0  # time of last client access to the camera  

58.     event = CameraEvent()  


60.     def __init__(self):  

61.         """Start the background camera thread if it isn't running yet."""  

62.         if BaseCamera.thread is None:  

63.             BaseCamera.last_access = time.time()  


65.             # start background frame thread  

66.             BaseCamera.thread = threading.Thread(target=self._thread)  

67.             BaseCamera.thread.start()  


69.             # wait until frames are available  

70.             while self.get_frame() is None:  

71.                 time.sleep(0)  


73.     def get_frame(self):  

74.         """Return the current camera frame."""  

75.         BaseCamera.last_access = time.time()  


77.         # wait for a signal from the camera thread  

78.         BaseCamera.event.wait()  

79.         BaseCamera.event.clear()  


81.         return BaseCamera.frame  


83.     @staticmethod  

84.     def frames():  

85.         """"Generator that returns frames from the camera."""  

86.         raise RuntimeError('Must be implemented by subclasses.')  


88.     @classmethod  

89.     def _thread(cls):  

90.         """Camera background thread."""  

91.         print('Starting camera thread.')  

92.         frames_iterator = cls.frames()  

93.         for frame in frames_iterator:  

94.             BaseCamera.frame = frame  

95.             BaseCamera.event.set()  # send signal to clients  

96.             time.sleep(0)  


98.             # if there haven't been any clients asking for frames in  

99.             # the last 10 seconds then stop the thread  

100.             if time.time() - BaseCamera.last_access > 10:  

101.                 frames_iterator.close()  

102.                 print('Stopping camera thread due to inactivity.')  

103.                 break  

104.         BaseCamera.thread = None  


When developing any function related with OpenCV in the following tutorial, you only need to include the respective file in the same folder with and and run in the Raspberry Pi command line.

Open a web browser on the device under the same LAN with the Raspberry Pi, enter the Raspberry Pi's IP address with the port :5000, as shown below:


11.5 Real-time video display on the web page

The video display in real time by web controller is implemented based on the OpenCV function as mentioned above. A web controller is a web interface to control the robot product to perform various actions and it can be applied on any device that is able to run a browser, including PC, mobile phones, tablets, etc.

If you've completed all installations based on the instructional document, it will be quite easy to open a web controller.

Check that your device is under the same LAN with the Raspberry Pi.

Obtain the Raspberry Pi's IP address.

If you terminate the Raspberry Pi auto-run program, you need to re-run the program, run the command:

sudo python3 adeept_rasptankpro/server/

Open a web browser (recommended to use Chrome in case of any possible incompatibility with other browsers), enter the Raspberry Pi's IP address with the port :5000, for instance:


Then the web controller will be loaded into the browser.


The video window on the top left corner shows the images from the Raspberry Pi camera. Modules on the web controller may vary from products. More details of the modules will be explained subsequently.