opencv将图片绕中心点旋转任意角度后嵌入另一张图片中心并记录相关坐标点位置信息

import cv2
import numpy as np
import math
import os
import pandas as pd
def __blend_imgquadbox(image1, image2, bbox,label, image_coordinate=None, scale=1.5):
    assert label.shape[0]==bbox.shape[0]
    h_image1, w_image1, _ = image1.shape
    h_image2, w_image2, _ = image2.shape
    if h_image1 <= h_image2 or w_image1 <= w_image2:
        image1 = cv2.resize(image1, dsize=(int(w_image2*scale), int(h_image2*scale)))
    h_image1, w_image1, _ = image1.shape
    center_x, center_y = w_image1 / 2, h_image1 / 2
    length_x = np.int0(center_x - w_image2 / 2)
    length_y = np.int0(center_y - h_image2 / 2)
    bbox[:,:,0] += length_x
    bbox[:,:,1] += length_y
    if image_coordinate is not None:
        image_coordinate[:, 0] += length_x
        image_coordinate[:, 1] += length_y
        cv2.fillPoly(image1, [image_coordinate], (0, 0, 0))
    image1[length_y:length_y + h_image2,length_x:length_x + w_image2] = cv2.add(image1[length_y:length_y + h_image2,length_x:length_x + w_image2], image2)
    return image1, bbox, label

def _resize_imgquadbox(image, bbox, label, image_coordinate=None, dsize=(800,600)):
    assert label.shape[0]==bbox.shape[0]
    h, w, _ = image.shape
    ratio_h, ratio_w = h / dsize[1], w / dsize[0]
    bbox[:,:4,0] = bbox[:,:4,0] // ratio_w
    bbox[:,:4,1] = bbox[:,:4,1] // ratio_h
    img = cv2.resize(image,dsize)
    if image_coordinate is None:
        return img, bbox
    image_coordinate[:,0] = image_coordinate[:, 0] // ratio_w
    image_coordinate[:,1] = image_coordinate[:, 1] // ratio_h
    return img, bbox, label, image_coordinate

def _rotate_imgquadbox(image, bbox, label, angle):
    assert label.shape[0] == bbox.shape[0]
    h, w, _ = image.shape
    cX, cY = w / 2, h / 2
    M = cv2.getRotationMatrix2D((cX, cY), -angle, 1.0)
    cos = np.abs(M[0, 0])
    sin = np.abs(M[0, 1])

    nw = int((h * sin) + (w * cos))
    nh = int((h * cos) + (w * sin))

    M[0, 2] += (nw / 2) - cX
    M[1, 2] += (nh / 2) - cY

    image_x1 = np.array([0, w, w, 0])
    image_y1 = np.array([h, h, 0, 0])#h - [0, 0, h, h]
    x1 = bbox[:,:4, 0]
    y1 = h - bbox[:,:4, 1]
    x2 = cX
    y2 = h - cY
    
    #由于图片坐标y值与欧几里得坐标正好相反，所以这里sin值默认取相反数（对比论文里的公式自己动手算一下就明白了）
    image_x = (image_x1 - x2)*math.cos(math.pi / 180.0 * -angle) - (image_y1 - y2)*math.sin(math.pi / 180.0 * -angle) + x2
    image_y = (image_x1 - x2)*math.sin(math.pi / 180.0 * -angle) + (image_y1 - y2)*math.cos(math.pi / 180.0 * -angle) + y2
    x = (x1 - x2)*math.cos(math.pi / 180.0 * -angle) - (y1 - y2)*math.sin(math.pi / 180.0 * -angle) + x2
    y = (x1 - x2)*math.sin(math.pi / 180.0 * -angle) + (y1 - y2)*math.cos(math.pi / 180.0 * -angle) + y2
    image_x = np.int0(image_x + (nw - w) / 2).flatten()
    image_y = np.int0(h - image_y + (nh - h) / 2).flatten()
    x = np.int0(x + (nw - w) / 2).flatten()
    y = np.int0(h - y + (nh - h) / 2).flatten()

    image_coordinate = np.array([[x,y] for x, y in zip(image_x, image_y)]).reshape(-1, 2)
    bbox = np.array([[x, y] for x, y in zip(x, y)]).reshape(bbox.shape)
    image = cv2.warpAffine(image, M, (nw, nh))
    return image, bbox, label, image_coordinate

def blend2img(img1, img2, bbox, label, angle):
    """
    :param img1: dst img
    :param img2: roi img
    :param bbox: quadbox in roi img(4 point represent one rectangle box in roi img)
    :param label: each label math one quadbox
    :param angle: clockwise rotate angle
    :return:
    """
    if isinstance(img1, np.ndarray):
        img1 = img1
    elif isinstance(img1, str):
        img1 = cv2.imread(img1)
    else:
        raise Exception("img must be ndarray or str(img path)", img1)

    if isinstance(img2, np.ndarray):
        img2 = img2
    elif isinstance(img2, str):
        img2 = cv2.imread(img2)
    else:
        raise Exception("img must be ndarray or str(img path)", img2)
    assert bbox.ndim==3
    assert bbox.shape[0] == label.shape[0]
    assert label.shape[1]==1
    image, bbox, label = __blend_imgquadbox(img1,*_resize_imgquadbox(*_rotate_imgquadbox(img2, bbox, label, angle)))
    return image, bbox, label


if __name__ == '__main__':

    imglist = os.listdir("zzzmydata/bbbox")
    expresslist = os.listdir("zzzmydata/express")
    orgincsvdata = pd.read_csv("zzzmydata/annotation.csv")

    xml_list = []
    for img in imglist:
        img1 = cv2.imread("zzzmydata/bbbox/{}".format(img))
        for ep in expresslist:
            img2 = cv2.imread("zzzmydata/express/{}".format(ep))
            bbox = np.array(orgincsvdata[orgincsvdata['filename'] == ep].loc[:,['x1','y1','x2','y2','x3','y3','x4','y4']].values)
            newbbox = bbox.reshape(bbox.shape[0],-1, 2)
            label = np.array(orgincsvdata[orgincsvdata['filename'] == ep].loc[:, ['class']].values)
            for angle in [-90,-75,-60,-45,-30,-15,0,15,30,45,60,75,90]:
                lastimage, lastbbox, lastlabel = blend2img(img1.copy(), img2.copy(), newbbox.copy(), label.copy(), angle)
                resultbbox = lastbbox.reshape(bbox.shape[0],8)
                for i in range(lastlabel.shape[0]):
                    value = ("{}_{}_{}".format(angle,ep.split('.')[0],img), resultbbox[i][0], resultbbox[i][1], resultbbox[i][2], resultbbox[i][3], resultbbox[i][4], resultbbox[i][5], resultbbox[i][6], resultbbox[i][7],lastlabel[i][0],angle)
                    xml_list.append(value)
                cv2.imwrite("newbarcodedata/train/{}_{}_{}".format(angle,ep.split('.')[0],img), lastimage)

    column_name = ['filename', 'x1', 'y1', 'x2', 'y2','x3','y3','x4','y4','class','angle']
    xml_df = pd.DataFrame(xml_list,columns=column_name)
    xml_df.to_csv(("newbarcodedata/train_ann.csv"),index=None)
    print('Successfully make mydataset.')

 参考论文：RRPN