根据之前视频防抖的代码，重新写了一套。能够输出缩放、旋转的信息。

代码

import cv2
import numpy as np
import argparse
import os
import re
import math
import matplotlib.pyplot as plt
import time

def show_image(image):
plt.figure(figsize=(15, 10))
plt.imshow(image)
plt.title('Feature Matching Results')
plt.show()

def process_transform(refer_kp, drone_kp, matches, drone_frame):
# 提取匹配点的坐标
refer_pts = np.float32([refer_kp[m.queryIdx].pt for m in matches]).reshape(-1, 1, 2)
drone_pts = np.float32([drone_kp[m.trainIdx].pt for m in matches]).reshape(-1, 1, 2)

复制代码

  # 计算单应性矩阵
  H, mask = cv2.findHomography(drone_pts, refer_pts, cv2.RANSAC, 5.0)
      
  if H is None:
      print(f"无法计算 {frame_files[i]} 的变换矩阵，使用上一帧的变换")
      # 如果无法计算，使用恒等变换
      H = np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1]])
      
  # 从单应性矩阵中提取变换信息
  scale        = np.sqrt(H[0,0]**2 + H[1,0]** 2)
  rotation_rad = math.atan2(H[1,0], H[0,0])
  rotation_deg = math.degrees(rotation_rad)
  x_offset     = H[0,2]
  y_offset     = H[1,2]
  print('transform: scale=%.2f, rotation_rad=%.2f, rotation_deg=%.2f, x_offset=%.2f, y_offset=%.2f' % (scale, rotation_rad, rotation_deg, x_offset, y_offset))
      
  # 应用变换矩阵进行修正
  # height, width  = drone_frame.shape[:2]
  # adjusted_frame = cv2.warpPerspective(drone_frame, H, (width, height))
  # cv2.imwrite(adjusted_frame_path, adjusted_frame)

def process_frame(frame):
# 读取第一帧作为初始基准帧
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)

复制代码

  # 检测基准帧的特征点和描述符
  kp, des = orb.detectAndCompute(gray, None)
  return frame, gray, kp, des

def map_match(reference, drone):

复制代码

  current_time = time.time()
  refer_frame, refer_gray, refer_kp, refer_des = process_frame(reference)
  drone_frame, drone_gray, drone_kp, drone_des = process_frame(drone)
      
  # 匹配特征点（当前帧与上一调整后的帧）
  matches = matcher.match(refer_des, drone_des)
  # 按匹配距离排序，取前100个最佳匹配
  matches = sorted(matches, key=lambda x: x.distance)[:10]
      
  process_transform(refer_kp, drone_kp, matches, drone_frame)

  print('cost time: %.2f ms' % ((time.time()-current_time)*1000))
  
  match_img = cv2.drawMatches(refer_frame, refer_kp,
                         drone_frame, drone_kp,
                         matches, None,
                         flags=cv2.DrawMatchesFlags_NOT_DRAW_SINGLE_POINTS)

  show_image(match_img)

初始化ORB特征检测器和匹配器

orb = cv2.ORB_create(1000)
matcher = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)

if name == "main":
for index in range(1, 10):
image1 = ('images/map-%02d.jpg' % ((index-1)*3))
image1 = cv2.imread(image1)
image2 = ('images/map-%02d.jpg' % ( index *3))
image2 = cv2.imread(image2)
map_match(image1, image2)

效果

准确的计算出旋转角度，90。

复制代码

transform: scale=0.72, rotation_rad=-1.57, rotation_deg=-89.81, x_offset=184.79, y_offset=737.42
cost time: 45.87 ms

地图影像匹配：基于特征匹配的视觉定位2，python

初始化ORB特征检测器和匹配器