[双目视觉] SGBM 算法应用（Python 版本）
最编程
2024-03-31 07:46:18
...
import numpy as np
import cv2
import random
import math
# 加载视频文件
capture = cv2.VideoCapture("./car.avi")
WIN_NAME = 'Deep disp'
cv2.namedWindow(WIN_NAME, cv2.WINDOW_AUTOSIZE)
# 读取视频
fps = 0.0
ret, frame = capture.read()
while ret:
# 开始计时
t1 = time.time()
# 是否读取到了帧，读取到了则为True
ret, frame = capture.read()
# 切割为左右两张图片
frame1 = frame[0:480, 0:640]
frame2 = frame[0:480, 640:1280]
# 将BGR格式转换成灰度图片，用于畸变矫正
imgL = cv2.cvtColor(frame1, cv2.COLOR_BGR2GRAY)
imgR = cv2.cvtColor(frame2, cv2.COLOR_BGR2GRAY)
# 重映射，就是把一幅图像中某位置的像素放置到另一个图片指定位置的过程。
# 依据MATLAB测量数据重建无畸变图片,输入图片要求为灰度图
img1_rectified = cv2.remap(imgL, left_map1, left_map2, cv2.INTER_LINEAR)
img2_rectified = cv2.remap(imgR, right_map1, right_map2, cv2.INTER_LINEAR)
# 转换为opencv的BGR格式
imageL = cv2.cvtColor(img1_rectified, cv2.COLOR_GRAY2BGR)
imageR = cv2.cvtColor(img2_rectified, cv2.COLOR_GRAY2BGR)
# ------------------------------------SGBM算法----------------------------------------------------------
# blockSize 深度图成块，blocksize越低，其深度图就越零碎，0<blockSize<10
# img_channels BGR图像的颜色通道，img_channels=3，不可更改
# numDisparities SGBM感知的范围，越大生成的精度越好，速度越慢，需要被16整除，如numDisparities
# 取16、32、48、64等
# mode sgbm算法选择模式，以速度由快到慢为：STEREO_SGBM_MODE_SGBM_3WAY、
# STEREO_SGBM_MODE_HH4、STEREO_SGBM_MODE_SGBM、STEREO_SGBM_MODE_HH。精度反之
# ------------------------------------------------------------------------------------------------------
blockSize = 8
img_channels = 3
stereo = cv2.StereoSGBM_create(minDisparity=1,
numDisparities=64,
blockSize=blockSize,
P1=8 * img_channels * blockSize * blockSize,
P2=32 * img_channels * blockSize * blockSize,
disp12MaxDiff=-1,
preFilterCap=1,
uniquenessRatio=10,
speckleWindowSize=100,
speckleRange=100,
mode=cv2.STEREO_SGBM_MODE_HH)
# 计算视差
disparity = stereo.compute(img1_rectified, img2_rectified)
# 归一化函数算法，生成深度图（灰度图）
disp = cv2.normalize(disparity, disparity, alpha=0, beta=255, norm_type=cv2.NORM_MINMAX, dtype=cv2.CV_8U)
# 生成深度图（颜色图）
dis_color = disparity
dis_color = cv2.normalize(dis_color, None, alpha=0, beta=255, norm_type=cv2.NORM_MINMAX, dtype=cv2.CV_8U)
dis_color = cv2.applyColorMap(dis_color, 2)
# 计算三维坐标数据值
threeD = cv2.reprojectImageTo3D(disparity, Q, handleMissingValues=True)
# 计算出的threeD，需要乘以16，才等于现实中的距离
threeD = threeD * 16
# 鼠标回调事件
cv2.setMouseCallback("depth", onmouse_pick_points, threeD)
#完成计时，计算帧率
fps = (fps + (1. / (time.time() - t1))) / 2
frame = cv2.putText(frame, "fps= %.2f" % (fps), (0, 40), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
cv2.imshow("depth", dis_color)
cv2.imshow("left", frame1)
cv2.imshow(WIN_NAME, disp) # 显示深度图的双目画面
# 若键盘按下q则退出播放
if cv2.waitKey(20) & 0xff == ord('q'):
break
# 释放资源
capture.release()
# 关闭所有窗口
cv2.destroyAllWindows()