双目视觉07_立体匹配Census算法

    import os
    import time
    import cv2 as cv
    import numpy as np
    import matplotlib.pyplot as plt
    from numba import jit
    
    # lres = census(limg[:, maxDisparity:w])
    # maxDisparity = 12  # 最大视差，也就是搜索范围，搜索范围越大耗时越长
    # window_size = 5  # 滑动窗口大小，窗口越大匹配成功难度越高
    # same = np.int64(window_size / 2)  # 窗口大小的一半向下取整，比如用来衡量窗口左上角与窗口中点的距离的x与y坐标
    
    
    @jit
    def census(img):
    size = np.shape(img)[0:2]
    h = size[0]
    w = size[1]
    census = np.zeros((h - 2 * same, w - 2 * same), dtype=np.int64)
    # census就是经过变换后的图像
    cp = img[same:h - same, same:w - same]
    # cp是变换中心点的坐标，0，0点作为左上角的窗口中，中心点就是(same,same)
    offsets = [(u, v) for v in range(window_size) for u in range(window_size) if not u == same == v]
    # offset是分别衡量某个点作为窗口左上角是否匹配成功，作为窗口左上角向左平移一个匹配是否成功，一直到作为窗口右下角匹配是否成功，都有衡量
    for u, v in offsets:
        census = (census << 1) | (img[v:v + h - 2 * same,
                                  u:u + w - 2 * same] >= cp)  # 这里是最为难懂的，请参考一下[我的另一篇博文]()，如果还有不懂请私信我或给我留言
    return census
    
    
    def hanming(limg, rimg, dispmap, Diffmap):
    size = np.shape(limg)[0:2]
    h = size[0]
    w = size[1]
    lres = census(limg[:, maxDisparity:w])
    rres = census(rimg[:, 0:w - maxDisparity])
    for x in range(0, np.shape(lres)[0]):
        for y in range(maxDisparity, np.shape(lres)[1]):
            for i in range(0, maxDisparity):
                dispmap[x, y, i] = '{:025b}'.format(((rres[x, y - i]) ^ (lres[x, y]))).count(
                    '0')  # 从当前位置开始，在右图中寻找和左图最接近的二进制串，从而完成匹配
    for x in range(0, np.shape(lres)[0]):
        for y in range(0, np.shape(lres)[1]):
            val = np.sort(dispmap[x, y, :])  # 排名
            val_id = np.argsort(dispmap[x, y, :])  # 找出最接近的
            Diffmap[x, y] = dispmap[x, y, :].argsort()[-1]
    
    
    
    
    
    # os.chdir(r'')
    limg = np.asanyarray(cv.imread('pic/left/left.png', cv.IMREAD_GRAYSCALE), dtype=np.int64)
    rimg = np.asanyarray(cv.imread('pic/right/right.png', cv.IMREAD_GRAYSCALE), dtype=np.int64)
    # img_size=np.shape(limg)
    size = np.shape(limg)[0:2]
    h = size[0]
    w = size[1]
    
    
    maxDisparity = 12  # 最大视差，也就是搜索范围，搜索范围越大耗时越长
    window_size = 5  # 滑动窗口大小，窗口越大匹配成功难度越高
    same = np.int64(window_size / 2)  # 窗口大小的一半向下取整，比如用来衡量窗口左上角与窗口中点的距离的x与y坐标
    lres = census(limg[:, maxDisparity:w])
    
    
    rres = np.zeros((np.shape(lres)[0], np.shape(lres)[1], maxDisparity), dtype=np.int64)
    dispmap = np.zeros((np.shape(lres)[0], np.shape(lres)[1], maxDisparity), dtype=np.int64)
    Diffmap = np.zeros((np.shape(lres)[0], np.shape(lres)[1]), dtype=np.int64)
    tic1 = time.time()
    hanming(limg, rimg, dispmap, Diffmap)
    plt.imshow(Diffmap[:, maxDisparity:])
    plt.show()
    print('用时:', time.time() - tic1)