【辅助驾驶】图像拼接[2]——普通图像拼接实现

一、效果

二、流程

1. 对每幅图进行特征点提取
2. 对对特征点进行匹配
3. 进行图像配准
4. 把图像拷贝到另一幅图像的特定位置
5. 对重叠边界进行特殊处理

三、代码

基于Python实现Sift特征提取，之后尽心匹配拼接

import cv2 as cv
import numpy as np
import os
import copy


def extractFeature(imgR, imgL, detector="sift", pointsNum=None):
    if detector == "sift":
        feature = cv.xfeatures2d.SIFT_create(pointsNum)
    elif detector == "surf":
        feature = cv.xfeatures2d.SURF_create(pointsNum)
    else:
        feature = cv.ORB_create(pointsNum)
        # kpR, desR = feature.detectAndCompute(imgR, None)
        # kpL, desL = feature.detectAndCompute(imgL, None)
        # bf = cv.BFMatcher(cv.NORM_HAMMING)
        # matches = bf.knnMatch(desR, desL, k=2)
    kpR, desR = feature.detectAndCompute(imgR, None)
    kpL, desL = feature.detectAndCompute(imgL, None)
    bf = cv.BFMatcher(cv.NORM_L2)
    matches = bf.knnMatch(desR, desL, k=2)
    return [matches, kpR, kpL]


def Optimize(imgL, warpR, imgTmp):
    imgRes = imgTmp
    alpha = 1
    rows, cols = np.where(warpR[:, :, 0] != 0)
    start = min(cols)
    width = imgL.shape[1] - start
    for i in range(imgL.shape[0]):
        for j in range(start, imgL.shape[1]):
            if warpR[i, j, :].all() == 0:
                alpha = 1
                # alpha = (width - (j - start)) / width
            else:
                alpha = (width - (j - start)) / width
                # print([j,alpha])
            imgRes[i, j, :] = imgL[i, j, :] * alpha + wrapR[i, j, :] * (1 - alpha)
    return imgRes


if __name__ == '__main__':
    detector = "sift"
    cwd = os.getcwd()
    imgR = cv.imread("./originImg/05.jpg")
    imgL = cv.imread("./originImg/06.jpg")
    imgL = imgL[:,0:235,:]
    for pNum in range(500, 10000, 500):
        # 提取初步特征点
        # imgR = imgR.astype(np.float32)
        # imgL = imgL.astype(np.float32)
        [matches, kpR, kpL] = extractFeature(imgR, imgL, detector=detector, pointsNum=pNum)
        good = [m for (m, n) in matches if m.distance < 0.75 * n.distance]  # 获取关键点的坐标
        src_pts = np.float32([kpR[m.queryIdx].pt for m in good]).reshape(-1, 1, 2)
        dst_pts = np.float32([kpL[m.trainIdx].pt for m in good]).reshape(-1, 1, 2)

        H, mask = cv.findHomography(src_pts, dst_pts, cv.RANSAC, 5.0)
        # cv.warpPerspective(img,h,(cols,rows))
        #print((imgR.shape[1] + imgL.shape[1]).type())
        wrapR = cv.warpPerspective(imgR, H, (imgR.shape[1] + imgL.shape[1], imgL.shape[0]))
        imgTmp = copy.deepcopy(wrapR)
        imgTmp[0:imgL.shape[0], 0:imgL.shape[1]] = imgL
        # cv.imshow('wrap2.jpg', imgTmp)
        # cv.imshow('wrap1.jpg', wrapR)

        rows, cols = np.where(imgTmp[:, :, 0] != 0)
        min_row, max_row = min(rows), max(rows) + 1
        min_col, max_col = min(cols), max(cols) + 1
        imgTmp = imgTmp[min_row:max_row, min_col:max_col, :]  # 去除黑色无用部分
        # 图像融合
        imgRes = Optimize(imgL, wrapR, imgTmp)

        # # 形态学处理,开运算
        # g = cv.getStructuringElement(cv.MORPH_RECT, (9, 9))
        # imgOpen = cv.morphologyEx(imgRes, cv.MORPH_OPEN, g)
        # res = abs(imgRes - imgTmp)
        resPath = cwd + "\\" + detector
        if not os.path.exists(resPath):
            os.mkdir(resPath)
        cv.imwrite(resPath + "\\" + "res" + str(pNum) + ".jpg", imgRes)
# cv.imshow('imgR.jpg', imgR)
# cv.imshow('imgL.jpg', imgL)
# cv.imshow('result.jpg', imgRes)
# # cv.imshow('res.jpg', res)
# cv.waitKey(0)
# cv.destroyAllWindows()

四、项目下载地址

Github：https://github.com/linghugoogle/ImageStitch