基于二值图的角点检测

简述

cv2.goodFeaturesToTrack()函数是用来跟踪检测图像中的角点

参数

• image: 输入图像，是八位的或者32位浮点型，单通道图像，所以有时候用灰度图 maxCorners:
  返回最大的角点数，是最有可能的角点数，如果这个参数不大于0，那么表示没有角点数的限制。
• qualityLevel:图像角点的最小可接受参数，质量测量值乘以这个参数就是最小特征值，小于这个数的会被抛弃。
• minDistance:返回的角点之间最小的欧式距离。
• mask: 检测区域。如果图像不是空的(它需要具有CV_8UC1类型和与图像相同的大小)，它指定检测角的区域。
• blockSize:用于计算每个像素邻域上的导数协变矩阵的平均块的大小。
• useHarrisDetector：选择是否采用Harris角点检测，默认是false. k: Harris检测的自由参数。

代码

import cv2
import numpy as np
import scipy.ndimage
import skimage.morphology
import os
import gdalTools


def good_feature_to_track(thin_mask, mask, out_name, save_path):
    """
     Apply the detector on the segmentation map to detect the road junctions as starting points for tracing.
    :param thin_mask: one-pixel width segmentation map
    :param mask: road segmentation map
    :param out_name: filename
    :param save_path: the directory of corner detection results
    :return:
    """
    # set a padding to avoid image edge corners
    padding_x = 64+5
    padding_y = 64

    corners = cv2.goodFeaturesToTrack(thin_mask, 100, 0.1, 10)
    corners = np.int0(corners)
    img = np.zeros((mask.shape[0], mask.shape[1], 3))
    img[:, :, 0] = mask
    img[:, :, 1] = mask
    img[:, :, 2] = mask
    corner_num = 0
    with open(save_path+out_name[:-4]+".txt", "w") as f:
        for i in corners:
            x, y = i.ravel()
            if x < padding_x or x > img.shape[0]-padding_x:
                continue
            if y < padding_y or y > img.shape[1]-padding_y:
                continue

            f.write("{},{}\n".format(x,y))
            cv2.circle(img, (x, y), 6, (0, 0, 255), -1)
            corner_num += 1
    print("total corners number:{}".format(corner_num))
    # cv2.imwrite(save_path+out_name[:-4]+'_with_corners.png', img)
    return img


def thin_image(mask_dir, filename):
    """
    Skeletonize the road segmentation map to a one-pixel width
    :param mask_dir: the directory of road segmentation map
    :param filename: the filename of road segmentation map
    :return: one-pixel width segmentation map
    """
    im = scipy.ndimage.imread(mask_dir + filename)
    im = im > 128
    selem = skimage.morphology.disk(2)
    im = skimage.morphology.binary_dilation(im, selem)
    im = skimage.morphology.thin(im)
    return im.astype(np.uint8) * 255


def mkdir(path):
    if not os.path.exists(path):
        os.mkdir(path)


if __name__ == '__main__':
    import matplotlib.pyplot as plt
    txt_dir = 'corners'
    mkdir(txt_dir)
    mask_dir = 'train_labels/'
    mask_filename = '000000011.tif'
    thin_img = thin_image(mask_dir, mask_filename)
    # mask = cv2.imread(mask_dir + mask_filename, 0)
    im_proj, im_geotrans, im_width, im_height, mask = gdalTools.read_img(mask_dir + mask_filename)
    img = good_feature_to_track(thin_img, mask, mask_filename, txt_dir)
    plt.subplot(121)
    plt.imshow(thin_img * 255)
    plt.subplot(122)
    plt.imshow(img)
    plt.show()