【计算机视觉】使用普通相机进行深度估计

问题描述：

深度估计是计算机视觉中的一个主要问题，特别是对于与增强现实、机器人甚至自动驾驶汽车相关的应用。

传统的 3D 传感器通常使用立体视觉、运动或结构光的投影。然而，这些传感器依赖于环境（阳光、纹理）或需要多个外围设备（相机、投影仪），这导致系统非常庞大。

最近，已经提出了几种基于深度学习的深度估计方法。这些方法使用单个视点（单个图像）并且通常优化参考深度图上的回归。

实现步骤：

1.图片下载https://github.com/PacktPublishing/Learning-OpenCV-4-Computer-Vision-with-Python-Third-Edition/blob/master/images/color1_small.jpg

​​​​​​https://github.com/PacktPublishing/Learning-OpenCV-4-Computer-Vision-with-Python-Third-Edition/blob/master/images/color2_small.jpg

2.我們為立體算法的幾個參數定義初始值

import numpy as np
import cv2 
minDisparity = 16 
numDisparities = 192 - minDisparity

blockSize = 5
uniquenessRatio = 1 
speckleWindowSize = 3 
speckleRange = 3 
disp12MaxDiff = 200
P1 = 600
P2 = 2400

3.使用這些參數，我們創建了 OpenCV 的 cv2.StereoSGBM 類的實例。

SGBM 代表半全局塊匹配，這是一種用於計算視差圖的算法。

stereo = cv2.StereoSGBM_create(
    minDisparity = minDisparity, 
    numDisparities = numDisparities, 
    blockSize = blockSize, 
    uniquenessRatio = uniquenessRatio, 
    speckleRange = speckleRange, 
    speckleWindowSize = speckleWindowSize, 
    disp12MaxDiff = disp12MaxDiff,
    P1 = P1,
    P2 = P2 
)

4.读取图片

imgl = cv2.imread('img/color1_small.jpeg')
imgr = cv2.imread('img/color2_small.jpeg')
imgl

5.我們希望提供幾個滑塊，使用戶能夠以交互方式調整計算視差圖的算法的參數。 每當用戶調整任何滑塊時，我們都會通過設置 StereoSGBM 實例的屬性來更新立體算法的參數，並通過調用 StereoSGBM 實例的計算方法重新計算視差圖。

def update(slidevalue =0):
    stereo.setBlockSize(
        cv2.getTrackbarPos('blockSize','Display')
    )

    stereo.setUniquenessRatio(
        cv2.getTrackbarPos(
            'uniquenessRatio', 'Disparity'
        )
    )
    stereo.setSpeckleWindowSize( 
        cv2.getTrackbarPos('speckleWindowSize', 'Disparity')
    )
    stereo.setSpeckleRange( 
        cv2.getTrackbarPos('speckleRange', 'Disparity')
    )
    stereo.setDisp12MaxDiff( 
        cv2.getTrackbarPos('disp12MaxDiff', 'Disparity')
    )
    disparity = stereo.compute(
        imgl, imgr).astype(np.float32) / 16.0

    cv2.imshow('left',imgl)
    cv2.imshow('right',imgr)
    cv2.imshow('Disparity',
        (disparity - minDisparity) / numDisparities)

6.创建窗口和滑块

cv2.namedWindow('Disparity') 
cv2.createTrackbar('blockSize', 'Disparity', blockSize, 21,update) 
cv2.createTrackbar('uniquenessRatio', 'Disparity',uniquenessRatio, 50, update) 
cv2.createTrackbar('speckleWindowSize', 'Disparity',speckleWindowSize, 200, update) 
cv2.createTrackbar('speckleRange', 'Disparity',speckleRange, 50, update) 
cv2.createTrackbar('disp12MaxDiff', 'Disparity',
                      disp12MaxDiff, 250, update)

7.运行查看

update()

#wait for the user to press any key
#meanwhile update() will be called anytime the user moves a slider
cv2.waitKey()

运行截图：

代码展示：

import numpy as np
import cv2 



minDisparity = 16 
numDisparities = 192 - minDisparity

blockSize = 5
uniquenessRatio = 1 
speckleWindowSize = 3 
speckleRange = 3 
disp12MaxDiff = 200
P1 = 600
P2 = 2400


stereo = cv2.StereoSGBM_create(
    minDisparity = minDisparity, 
    numDisparities = numDisparities, 
    blockSize = blockSize, 
    uniquenessRatio = uniquenessRatio, 
    speckleRange = speckleRange, 
    speckleWindowSize = speckleWindowSize, 
    disp12MaxDiff = disp12MaxDiff,
    P1 = P1,
    P2 = P2 
)


imgl = cv2.imread('img/color1_small.jpeg')
imgr = cv2.imread('img/color2_small.jpeg')



def update(slidevalue =0):
    stereo.setBlockSize(
        cv2.getTrackbarPos('blockSize','Display')
    )

    stereo.setUniquenessRatio(
        cv2.getTrackbarPos(
            'uniquenessRatio', 'Disparity'
        )
    )
    stereo.setSpeckleWindowSize( 
        cv2.getTrackbarPos('speckleWindowSize', 'Disparity')
    )
    stereo.setSpeckleRange( 
        cv2.getTrackbarPos('speckleRange', 'Disparity')
    )
    stereo.setDisp12MaxDiff( 
        cv2.getTrackbarPos('disp12MaxDiff', 'Disparity')
    )
    disparity = stereo.compute(
        imgl, imgr).astype(np.float32) / 16.0

    cv2.imshow('left',imgl)
    cv2.imshow('right',imgr)
    cv2.imshow('Disparity',
        (disparity - minDisparity) / numDisparities)





cv2.namedWindow('Disparity') 
cv2.createTrackbar('blockSize', 'Disparity', blockSize, 21,update) 
cv2.createTrackbar('uniquenessRatio', 'Disparity',uniquenessRatio, 50, update) 
cv2.createTrackbar('speckleWindowSize', 'Disparity',speckleWindowSize, 200, update) 
cv2.createTrackbar('speckleRange', 'Disparity',speckleRange, 50, update) 
cv2.createTrackbar('disp12MaxDiff', 'Disparity',
                      disp12MaxDiff, 250, update)




update()


#wait for the user to press any key
#meanwhile update() will be called anytime the user moves a slider
cv2.waitKey()

参考：

《Learning OpenCV 4 Computer Vision with Python 3 - Third Edition》