MaixCAM MaixPy Use OpenCV

Introduction

For MaixCAM, since it uses Linux and the performance can support using the Python version of OpenCV, you can use the cv2 module directly in addition to the maix module.

The examples in this article and more can be found in MaixPy/examples/vision/opencv.

Note that OpenCV functions are basically CPU-calculated. If you can use maix modules, try not to use OpenCV, because many maix functions are hardware-accelerated.

Converting between Numpy/OpenCV and maix.image.Image Formats

You can convert maix.image.Image object to a numpy array, which can then be used by libraries such as numpy and opencv:

from maix import image, time, display, app

disp = display.Display()

while not app.need_exit():
    img = image.Image(320, 240, image.Format.FMT_RGB888)
    img.draw_rect(0, 0, 100, 100, image.COLOR_RED, thickness=-1)
    t = time.ticks_ms()
    img_bgr = image.image2cv(img, ensure_bgr=True, copy=True)
    img2   = image.cv2image(img_bgr, bgr=True, copy=True)
    print("time:", time.ticks_ms() - t)
    print(type(img_bgr), img_bgr.shape)
    print(type(img2), img2)
    print("")
    disp.show(img2)

The previous program is slower because each conversion involves a memory copy. Below is an optimized version for better performance. However, it is not recommended to use this unless you are aiming for extreme speed, as it is prone to errors:

from maix import image, time, display, app

disp = display.Display()

while not app.need_exit():
    img = image.Image(320, 240, image.Format.FMT_RGB888)
    img.draw_rect(0, 0, 100, 100, image.COLOR_RED, thickness=-1)

    t = time.ticks_ms()
    img_rgb = image.image2cv(img, ensure_bgr=False, copy=False)
    img2 = image.cv2image(img_rgb, bgr=False, copy=False)
    print("time:", time.ticks_ms() - t)
    print(type(img_rgb), img_rgb.shape)
    print(type(img2), img2)

    disp.show(img2)

• In img_rgb = image.image2cv(img, ensure_bgr=False, copy=False), img_rgb directly uses the data from img without creating a memory copy. Note that the obtained img_rgb is an RGB image. Since OpenCV APIs assume the image is BGR, you need to be careful when using OpenCV APIs to process the image. If you are not sure, set ensure_bgr to True.
• In img2 = image.cv2image(img_rgb, bgr=False, copy=False), setting copy to False means img2 directly uses the memory of img_rgb without creating a new memory copy, resulting in faster performance. However, be cautious because img_rgb must not be destroyed before img2 finishes using it; otherwise, the program will crash.
• Note that since memory is borrowed, modifying the converted image will also affect the original image.

Load an Image

import cv2

file_path = "/maixapp/share/icon/detector.png"
img = cv2.imread(file_path)
print(img)

Since the cv2 module is quite large, import cv2 may take some time.

Display Image on Screen

To display an image on the screen, convert it to a maix.image.Image object and then use display to show it:

from maix import display, image, time
import cv2

disp = display.Display()

file_path = "/maixapp/share/icon/detector.png"
img = cv2.imread(file_path)

img_show = image.cv2image(img)
disp.show(img_show)

while not app.need_exit():
    time.sleep(1)

Use OpenCV Functions

For example, edge detection:

Based on the code above, use the cv2.Canny function:

from maix import image, display, app, time
import cv2

file_path = "/maixapp/share/icon/detector.png"
img0 = cv2.imread(file_path)

disp = display.Display()

while not app.need_exit():
    img = img0.copy()

    # canny method
    t = time.ticks_ms()
    edged = cv2.Canny(img, 180, 60)
    t2 = time.ticks_ms() - t

    # show by maix.display
    t = time.ticks_ms()
    img_show = image.cv2image(edged)
    print(f"edge time: {t2}ms, convert time: {time.ticks_ms() - t}ms")
    disp.show(img_show)

Use Camera

On a PC, we use OpenCV's VideoCapture class to read from the camera. For MaixCAM, OpenCV does not support this directly, so we use the maix.camera module to read from the camera and then use it with OpenCV.

Convert a maix.image.Image object to a numpy.ndarray object using the image.image2cv function:

from maix import image, display, app, time, camera
import cv2

disp = display.Display()
cam = camera.Camera(320, 240)

while not app.need_exit():
    img = cam.read()

    # convert maix.image.Image object to numpy.ndarray object
    t = time.ticks_ms()
    img = image.image2cv(img)
    print("time: ", time.ticks_ms() - t)

    # canny method
    edged = cv2.Canny(img, 180, 60)

    # show by maix.display
    img_show = image.cv2image(edged)
    disp.show(img_show)