（Python代码）通过视差图获取图片中不同物体的高度比

1、原理

 可以推出不同物体的高度比可以通过如下公式得到：

 

  是左边指定物体的高度（图片中尺寸）

 是右边指定物体的高度（图片中尺寸）

  是左边指定物体的平均视差值

  是右边指定物体的平均视差值

 

2、代码逻辑框图

3、代码详解

import multiprocessing as mp
import numpy as np
import cv2
import time

#读取图像
img5_L = cv2.imread('relative_height/3_a.jpg')
img5_R = cv2.imread('relative_height/3_b.jpg')
img_L=img5_L
img_R=img5_R


#功能：获取视差图
#输入：两张图片
#输出：视差图
def disparity(img_L,img_R):
    stereo = cv2.StereoSGBM_create(numDisparities=48,blockSize=3)
    disparity = stereo.compute(img_L, img_R)
    return disparity

rect_L = (310,36,521,759) #起点x,y, 增加至为w,h
rect_R = (835,70,590,941)

#功能：截图
#输入：图片，截图范围
#输出：截图
def segment(img,rect):
    mask = np.zeros(img.shape[:2],np.uint8)
    bgdModel = np.zeros((1,65),np.float64)
    fgdModel = np.zeros((1,65),np.float64)
    cv2.grabCut(img,mask,rect,bgdModel,fgdModel,20,cv2.GC_INIT_WITH_RECT)
    mask2 = np.where((mask==2)|(mask==0),0,1).astype('uint8')
    img = img*mask2[:,:,np.newaxis]
    return img

#功能：构建高斯核sieze和sigma
#输入：size,sigma
#输出：二维数组
def gaussian_kernel_2d(kernel_size, sigma):
    kx = cv2.getGaussianKernel(kernel_size, sigma)
    ky = cv2.getGaussianKernel(kernel_size, sigma)
    return np.dot(kx, np.transpose(ky))


#功能：彩色图转成灰色图
#输入：彩色图img
#输出：灰色图gray_img
def color_to_gray(img):
    height = img.shape[0]
    width = img.shape[1]
    channels = img.shape[2]
    value = [0] * 3
    gray_img = np.zeros([height, width], np.uint8)
    for row in range(height):
        for column in range(width):
            for chan in range(channels):
                value[chan] = img[row, column, chan]
            R = value[2]
            G = value[1]
            B = value[0]
            # new_value = 0.2989 * R + 0.5870 * G + 0.1140 * B
            new_value = 0.2989 * R + 0.5870 * G + 0.1140 * B  # 转为灰度像素
            gray_img[row, column] = new_value
    return gray_img

#功能：获取截图的高度（像素之差）
#输入：img
#输出：高度
def object_height(img):
    point1=(0,0)
    point2 = (0, 0)
    height1=0
    height2=0
    for row in range(len(img)):
        for col in range(len(img[row])):
            if img[row][col] != 0:
                height1=row
                point1=(row,col)
                break

    for row in range(len(img)-1,-1,-1):
        for col in range(len(img[row])):
            if img[row][col] != 0:
                height2=row
                point2 = (row, col)
                break
    height=abs(height2-height1)
    return point1,point2,height





# 功能：求平均视差值
# 输入：抠图img,视差图disparity
# 输出：平均视差值
def average_disparity_value(img,disparity):
    n=0
    value=0
    for row in range(len(img)):
        for col in range(len(img[row])):
            if img[row][col] != 0:
                n +=1
                value+=disparity[row][col]
    return value/n



#功能：估计图片不同物体的实际高度比
#输入：图片img_L和img_R
#输出：图片中不同物体的实际高度比
def height_ratio(img_L,img_R):
    t_start = time.clock()
    print('compute disparity....')
    disparity1=disparity(img_L,img_R)
    disparity1 = cv2.filter2D(disparity1, -1, gaussian_kernel_2d(16, 60))
    print('object segment...')
    pool = mp.Pool(processes=mp.cpu_count() - 1) #并行处理设置
    res1=pool.apply_async(segment,(img_L, rect_L))#并行处理
    seg_L=res1.get()
    res2=pool.apply_async(segment, (img_R, rect_R))#并行处理
    seg_R = res2.get()
    print('color_to_gray...')
    seg_L=color_to_gray(seg_L)
    seg_R = color_to_gray(seg_R)
    print('computer object height...')
    point1_L,point2_L,height_L=object_height(seg_L)
    point1_R, point2_R, height_R = object_height(seg_R)
    d_L=average_disparity_value(seg_L, disparity1)
    d_R = average_disparity_value(seg_R, disparity1)
    t_end = time.clock()
    print('数据处理用时', t_end - t_start)
    ratio= height_L*d_R/(height_R*d_L)
    return ratio

if __name__ == '__main__':
    hei_ratio5=height_ratio(img_L,img_R)
    print(hei_ratio5)