Grass boundary detection when not 100% green

asked 2019-02-14 10:50:56 -0500

First of all, thank you so much for all the useful information on this forum. I managed to put together a python script to detect the lawn boundary for an autonomous lawn mower project I am working on. It can detect the boundary fine if the lawn is 100% green (below). image description

However, I encounter an issue if the lawn is not 100% green. This mage below has some patches of other colors, causing the boundary to be broken (below). image description

Please let me know if you have any suggestion on how I can fine-tune the script to better detect the boundary for the lawn in the second photo. Original image is here. image description.

My script is as follows:

    #RGB to HSV
    img_hsv = cv2.cvtColor(image, cv2.COLOR_RGB2HSV)

    #green mask
    lower_green = np.array( [40,40,40], dtype = "uint8")
    upper_green = np.array( [70,255,255], dtype = "uint8")
    mask_green = cv2.inRange(img_hsv, lower_green, upper_green)
    output = cv2.bitwise_and(image, image, mask=mask_green)

    #gaussian blur
    kernel_size = 3
    gauss = gaussian_blur(mask_green,kernel_size)

    kernel = np.ones((kernel_size*2,kernel_size*2),np.uint8)
    dilation_image = cv2.dilate(mask_green, kernel, iterations=1)

    #morph close
    closing = cv2.morphologyEx(dilation_image, cv2.MORPH_CLOSE, kernel)

    #remove small blobs
    nb_components, output, stats, centroids = cv2.connectedComponentsWithStats(closing, connectivity=8)
    #connectedComponentswithStats yields every separated component with information on each of them, such as size
    sizes = stats[1:, -1]; nb_components = nb_components - 1

    min_size = 150  #num pixels

    img2 = np.zeros((output.shape))
    for i in range(0, nb_components):
        if sizes[i] >= min_size:
            img2[output == i + 1] = 255

edit retag flag offensive close merge delete


follow tag grass

LBerger gravatar imageLBerger ( 2019-02-14 10:59:01 -0500 )edit

Thanks, LBerger. I managed to optimize my script to detect the boundary in the photo in the original post. Then I took a few actual photos of my lawn. The script performed miserably because the grass has shades of brown and yellow, e.g. here and here. It seems that the script also needs to do pattern detection. Please let me know if you have any suggestions on pattern detection.

model_leaf gravatar imagemodel_leaf ( 2019-03-17 12:27:44 -0500 )edit