好的！下面是一个基于Python的双目测距实现的基本流程说明以及示例代码框架，包括相机标定、立体校正和深度计算。 --- ### 双目测距简介 双目视觉是一种模拟人类双眼感知深度的方式。通过两个摄像头从不同角度拍摄同一场景，并利用视差原理（即左右图像对应点的位置差异），可以估算目标物体的距离。 以下是关键步骤： 1. **相机标定**：获取内参矩阵和畸变系数。 2. **立体校正**：对两幅图片进行极线矫正，使得匹配更容易。 3. **特征匹配**：找到左图和右图中的对应点并计算像素级视差。 4. **距离计算**：结合基线长度和焦距信息，将视差转化为实际距离。 --- ### 示例代码 #### 1. 相机标定 ```python import cv2 import numpy as np import glob # 标定棋盘格尺寸 (内部角点数) chessboard_size = (9, 6) # 准备对象点坐标 z=0 的平面 objp = np.zeros((np.prod(chessboard_size), 3), dtype=np.float32) objp[:, :2] = np.mgrid[0:chessboard_size[0], 0:chessboard_size[1]].T.reshape(-1, 2) # 存储对象点和图像点 objpoints_left = [] # 左摄像机的对象点 imgpoints_left = [] # 左摄像机的图像点 objpoints_right = [] imgpoints_right = [] # 加载标定图像路径 images_left = glob.glob('calibration/left/*.jpg') images_right = glob.glob('calibration/right/*.jpg') for img_path in images_left: img = cv2.imread(img_path) gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) ret, corners = cv2.findChessboardCorners(gray, chessboard_size, None) if ret: objpoints_left.append(objp) imgpoints_left.append(corners) for img_path in images_right: img = cv2.imread(img_path) gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) ret, corners = cv2.findChessboardCorners(gray, chessboard_size, None) if ret: objpoints_right.append(objp) imgpoints_right.append(corners) # 单独标定每一台相机 ret_left, mtx_left, dist_left, rvecs_left, tvecs_left = cv2.calibrateCamera( objpoints_left, imgpoints_left, gray.shape[::-1], None, None) ret_right, mtx_right, dist_right, rvecs_right, tvecs_right = cv2.calibrateCamera( objpoints_right, imgpoints_right, gray.shape[::-1], None, None) print("左相机内参:\n", mtx_left) print("右相机内参:\n", mtx_right) ``` --- #### 2. 立体校正与视差图生成 ```python # 获取立体标定参数 flags = cv2.CALIB_FIX_INTRINSIC stereocalib_criteria = (cv2.TERM_CRITERIA_MAX_ITER + cv2.TERM_CRITERIA_EPS, 100, 1e-5) ret, _, _, _, _, R, T, E, F = cv2.stereoCalibrate( objpoints_left, imgpoints_left, imgpoints_right, mtx_left, dist_left, mtx_right, dist_right, gray.shape[::-1], criteria=stereocalib_criteria, flags=flags ) # 计算校正映射 R1, R2, P1, P2, Q, validPixROI1, validPixROI2 = cv2.stereoRectify( mtx_left, dist_left, mtx_right, dist_right, gray.shape[::-1], R, T ) mapx_l, mapy_l = cv2.initUndistortRectifyMap(mtx_left, dist_left, R1, P1, gray.shape[::-1], cv2.CV_32FC1) mapx_r, mapy_r = cv2.initUndistortRectifyMap(mtx_right, dist_right, R2, P2, gray.shape[::-1], cv2.CV_32FC1) # 应用校正到输入图像 left_img = cv2.imread('test/left.jpg', 0) right_img = cv2.imread('test/right.jpg', 0) rectified_left = cv2.remap(left_img, mapx_l, mapy_l, cv2.INTER_LINEAR) rectified_right = cv2.remap(right_img, mapx_r, mapy_r, cv2.INTER_LINEAR) # 创建视差计算器 window_size = 3 min_disp = 16 num_disparities = 112 - min_disp blockSize = 7 stereo = cv2.StereoSGBM_create( minDisparity=min_disp, numDisparities=num_disparities, blockSize=blockSize, uniquenessRatio=10, speckleWindowSize=100, speckleRange=32, disp12MaxDiff=1, P1=8 * 3 * window_size ** 2, P2=32 * 3 * window_size ** 2 ) disparity = stereo.compute(rectified_left, rectified_right).astype(np.float32) / 16.0 cv2.imshow('Disparity Map', disparity / disparity.max()) cv2.waitKey(0) ``` --- #### 3. 距离计算公式 假设已知摄像头之间的基线长度 `baseline` 和焦距 `f`，则可以根据视差 `d` 计算距离 `Z`： \[ Z = \frac{baseline \times f}{d} \] 其中： - \( baseline \): 摄像头间距（单位为米） - \( f \): 镜头焦距（通常由标定获得） - \( d \): 像素级视差值 ---