人机交互实现图像全景拼接

注意

该代码只针对这两张图片进行的实现，对于其他的图像拼接请移步

opencv实现全景图像拼接及结果去黑_鸭鸭爱吃辣的博客-CSDN博客

事例图片

                       

                    

导入必要包

import cv2
import os
import math
import numpy as np

定义人机交互函数

def on_EVENT_LBUTTONDOWN(event, x, y, flags, param):
    global input_list
    a = len(input_list)
    if event == cv2.EVENT_LBUTTONDOWN:
        input_list.append([x,y])
        cv2.circle(img, (x, y), 3, (255, 0, 0), thickness = -1)
        cv2.putText(img, str(a+1), (x, y), cv2.FONT_HERSHEY_PLAIN,
                    1.0, (0,255,0), thickness = 1)
        cv2.imshow("window 2", img)

后面根据图片中相同的点集来计算图片之间的距离差值（由于点击误差太大，给设定了30像素的误差）

dataset_dir = 'photo' #输入文件夹
image_filenames = [(os.path.join(dataset_dir, x))
                    for x in os.listdir(dataset_dir)] #path为图片地址

all_list = []
for path in image_filenames:
    img0 = cv2.imread(path)
    img = img0.copy()

    input_list = list()

    cv2.namedWindow("window 1", flags=cv2.WINDOW_NORMAL | cv2.WINDOW_FREERATIO)
    cv2.imshow("window 1", img0)

    cv2.namedWindow("window 2", flags=cv2.WINDOW_NORMAL | cv2.WINDOW_FREERATIO)
    cv2.imshow("window 2", img)
    cv2.setMouseCallback("window 2", on_EVENT_LBUTTONDOWN)
    while (1):
        if cv2.waitKey(0):  # 按ESC键，循环终止，释放所有窗口资源
            break

    cv2.destroyAllWindows()  # 关闭窗口
    all_list+=[input_list[:4]] #截取前4个相同的点来作为参考（要按照顺序来进行）

img1 = cv2.imread('./photo/1.png')
img2 = cv2.imread('./photo/2.png')

EX = (all_list[0][0][0] + all_list[0][1][0] + all_list[0][1][0] + all_list[0][1][0])//4 - (all_list[1][0][0] +all_list[1][1][0] + all_list[1][2][0] + all_list[1][3][0])//4

tmp = img2[:,-EX+30:,:] #30为误差值

此时的tmp为需要拼接进img1的部分，然后再计算他所需旋转的角度即可，选取2点计算

img0 = cv2.imread('./photo/1.png')
img = img0.copy()

input_list = list()

cv2.namedWindow("window 1", flags=cv2.WINDOW_NORMAL | cv2.WINDOW_FREERATIO)
cv2.imshow("window 1", img0)

cv2.namedWindow("window 2", flags=cv2.WINDOW_NORMAL | cv2.WINDOW_FREERATIO)
cv2.imshow("window 2", img)
cv2.setMouseCallback("window 2", on_EVENT_LBUTTONDOWN)
while (1):
    if cv2.waitKey(0):  # 按ESC键，循环终止，释放所有窗口资源
        break

cv2.destroyAllWindows()  # 关闭窗口
list1 = input_list[:2]

img0 = cv2.imread('./photo/2.png')
img = img0.copy()

input_list = list()

cv2.namedWindow("window 1", flags=cv2.WINDOW_NORMAL | cv2.WINDOW_FREERATIO)
cv2.imshow("window 1", img0)

cv2.namedWindow("window 2", flags=cv2.WINDOW_NORMAL | cv2.WINDOW_FREERATIO)
cv2.imshow("window 2", img)
cv2.setMouseCallback("window 2", on_EVENT_LBUTTONDOWN)
while (1):
    if cv2.waitKey(0):  # 按ESC键，循环终止，释放所有窗口资源
        break

cv2.destroyAllWindows()  # 关闭窗口
list2 = input_list[:2]
#计算角度
jiaodu = math.atan2(list2[0][1]-list1[0][1],list2[0][0]-list1[0][0])

后面将tmp选择即可，最后接入img1即可

M = cv2.getRotationMatrix2D((EX-1,0),-jiaodu,1)

dst = cv2.warpAffine(tmp, M, (len(tmp[1]),389))
fin = np.hstack((img1, dst))
cv2.imshow('jieguo',fin)
cv2.imwrite('./out1/result.png'.format(len(fin)), fin)
cv2.waitKey(0)
cv2.destroyAllWindows()

结果 

总结

人机交互手搓误差太大，不如自动实现