data augmentation for object detecting目标检测xml文件扩增（旋转实现）

1. 背景描述：

在利用CNN做目标检测时，数据量不足时，旋转源图像进行数据的扩充。

例：
源图像如下图所示：

标记所得xml文件中目标信息如下：

<object>        <name>airplane</name>        <pose>Unspecified</pose>        <truncated>0</truncated>        <difficult>0</difficult>        <bndbox>            <xmin>431</xmin>            <ymin>367</ymin>            <xmax>607</xmax>            <ymax>453</ymax>        </bndbox>    </object>    <object>        <name>airplane</name>        <pose>Unspecified</pose>        <truncated>0</truncated>        <difficult>0</difficult>        <bndbox>            <xmin>570</xmin>            <ymin>419</ymin>            <xmax>768</xmax>            <ymax>512</ymax>

想要将源图像旋转任意角度，相对应xml文件中的bndbox信息则需要更新。

2. 思路：

参考博客(http://blog.csdn.net/u014540717/article/details/53301195)

找到原图中标记方框的四个边中点坐标，计算其旋转后的坐标位置，然后利用cv2.boundingRect函数找到四个新坐标的外接矩形作为新的xml文件中的bndbox值写入。

3. 代码实现过程：

# coding:utf-8# Copyright@hitzym, Dec,09,2017 at HIT# blog:http://blog.csdn.net/yinhuan1649/article/category/7330626import cv2import mathimport numpy as npimport xml.etree.ElementTree as ETimport osdef rotate_image(src, angle, scale=1):    w = src.shape[1]    h = src.shape[0]    # 角度变弧度    rangle = np.deg2rad(angle)  # angle in radians    # now calculate new image width and height    nw = (abs(np.sin(rangle) * h) + abs(np.cos(rangle) * w)) * scale    nh = (abs(np.cos(rangle) * h) + abs(np.sin(rangle) * w)) * scale    # ask OpenCV for the rotation matrix    rot_mat = cv2.getRotationMatrix2D((nw * 0.5, nh * 0.5), angle, scale)    # calculate the move from the old center to the new center combined    # with the rotation    rot_move = np.dot(rot_mat, np.array([(nw - w) * 0.5, (nh - h) * 0.5, 0]))    # the move only affects the translation, so update the translation    # part of the transform    rot_mat[0, 2] += rot_move[0]    rot_mat[1, 2] += rot_move[1]    dst = cv2.warpAffine(src, rot_mat, (int(math.ceil(nw)), int(math.ceil(nh))), flags=cv2.INTER_LANCZOS4)    # 仿射变换    return dst# 对应修改xml文件def rotate_xml(src, xmin, ymin, xmax, ymax, angle, scale=1.):    w = src.shape[1]    h = src.shape[0]    rangle = np.deg2rad(angle)  # angle in radians    # now calculate new image width and height    # 获取旋转后图像的长和宽    nw = (abs(np.sin(rangle)*h) + abs(np.cos(rangle)*w))*scale    nh = (abs(np.cos(rangle)*h) + abs(np.sin(rangle)*w))*scale    # ask OpenCV for the rotation matrix    rot_mat = cv2.getRotationMatrix2D((nw*0.5, nh*0.5), angle, scale)    # calculate the move from the old center to the new center combined    # with the rotation    rot_move = np.dot(rot_mat, np.array([(nw-w)*0.5, (nh-h)*0.5,0]))    # the move only affects the translation, so update the translation    # part of the transform    rot_mat[0, 2] += rot_move[0]    rot_mat[1, 2] += rot_move[1]                                   # rot_mat是最终的旋转矩阵    # point1 = np.dot(rot_mat, np.array([xmin, ymin, 1]))          #这种新画出的框大一圈    # point2 = np.dot(rot_mat, np.array([xmax, ymin, 1]))    # point3 = np.dot(rot_mat, np.array([xmax, ymax, 1]))    # point4 = np.dot(rot_mat, np.array([xmin, ymax, 1]))    point1 = np.dot(rot_mat, np.array([(xmin+xmax)/2, ymin, 1]))   # 获取原始矩形的四个中点，然后将这四个点转换到旋转后的坐标系下    point2 = np.dot(rot_mat, np.array([xmax, (ymin+ymax)/2, 1]))    point3 = np.dot(rot_mat, np.array([(xmin+xmax)/2, ymax, 1]))    point4 = np.dot(rot_mat, np.array([xmin, (ymin+ymax)/2, 1]))    concat = np.vstack((point1, point2, point3, point4))            # 合并np.array    # 改变array类型    concat = concat.astype(np.int32)    rx, ry, rw, rh = cv2.boundingRect(concat)                        #rx,ry,为新的外接框左上角坐标，rw为框宽度，rh为高度，新的xmax=rx+rw,新的ymax=ry+rh    return rx, ry, rw, rh# 使图像旋转60,90,120,150,210,240,300度xmlpath = './xml/'          #源图像路径imgpath = './imgs/'         #源图像所对应的xml文件路径rotated_imgpath = './rotatedimg/'rotated_xmlpath = './rotatedxml/'for angle in (60, 90, 120, 150, 180, 210, 240, 300):    for i in os.listdir(imgpath):        a, b = os.path.splitext(i)                            #分离出文件名a        img = cv2.imread(imgpath + a + '.jpg')        rotated_img = rotate_image(img,angle)        cv2.imwrite(rotated_imgpath + a + '_'+ str(angle) +'d.jpg',rotated_img)        print str(i) + ' has been rotated for '+ str(angle)+'°'        tree = ET.parse(xmlpath + a + '.xml')        root = tree.getroot()        for box in root.iter('bndbox'):            xmin = float(box.find('xmin').text)            ymin = float(box.find('ymin').text)            xmax = float(box.find('xmax').text)            ymax = float(box.find('ymax').text)            x, y, w, h = rotate_xml(img, xmin, ymin, xmax, ymax, angle)            # cv2.rectangle(rotated_img, (x, y), (x+w, y+h), [0, 0, 255], 2)   #可在该步骤测试新画的框位置是否正确            # cv2.imshow('xmlbnd',rotated_img)            # cv2.waitKey(200)            box.find('xmin').text = str(x)            box.find('ymin').text = str(y)            box.find('xmax').text = str(x+w)            box.find('ymax').text = str(y+h)        tree.write(rotated_xmlpath + a + '_'+ str(angle) +'d.xml')        print str(a) + '.xml has been rotated for '+ str(angle)+'°'

4. 测试旋转结果

将xml中的bounding box 显示在图片上用来测试旋转后结果是否正确

注：
- xml文件需要和其对应的jpg文件文件名一样
- e.g. monkey001.jpg 对应 monkey001.xml
- 上代码

# coding:utf-8# Copyright@hitzym, Dec,09,2017 at HIT# blog:http://blog.csdn.net/yinhuan1649/article/category/7330626import cv2import xml.etree.ElementTree as ETimport osimgpath = './testimgs/'          #旋转后的图像路径xmlpath = './testxml/'           #旋转后的xml文件路径for img in os.listdir(imgpath):    a, b = os.path.splitext(img)    img = cv2.imread(imgpath + a +'.jpg')    tree = ET.parse(xmlpath + a + '.xml')    root = tree.getroot()    for box in root.iter('bndbox'):        x1 = int(box.find('xmin').text)        y1 = int(box.find('ymin').text)        x2 = int(box.find('xmax').text)        y2 = int(box.find('ymax').text)        cv2.rectangle(img,(x1,y1),(x2, y2), [0,255,0], 2)    cv2.imshow("test", img)    # cv2.waitKey(1000)    if 1 == cv2.waitKey(0):        pass

原图：
结果图：
这是旋转60°的结果图

主要参考了博客(http://blog.csdn.net/u014540717/article/details/53301195)

稍有改动

感谢！