#
#作者:韦访
#博客:https://blog.csdn.net/rookie_wei
#微信:1007895847
#添加微信的备注一下是CSDN的
#欢迎大家一起学习
#
1、概述
前面几讲,我们已经分别实现了车牌检测和车牌号文本识别,现在就来将它们整合在一起进行完整的车牌识别。
环境配置:
操作系统:Ubuntu 64位
显卡:GTX 1080ti
Python:Python3.7
TensorFlow:2.3.0
2、车牌检测
首先需要做的是车牌检测,先导入所有要检测的图片,
'''
导入图片数据
'''
def get_images(data_path):
files = []
for ext in ['jpg', 'png', 'jpeg']:
files.extend(glob.glob(os.path.join(data_path, '*.{}'.format(ext))))
return files
接着导入并加载EAST模型,
'''
创建和加载east模型
'''
def east_create_and_load_model():
# 创建模型
# load trained model
model = EAST_model()
ckpt = tf.train.Checkpoint(step=tf.Variable(0), model=model)
latest_ckpt = tf.train.latest_checkpoint(FLAGS.east_model_path)
# 加载模型
if latest_ckpt:
ckpt.restore(latest_ckpt)
print('global_step : {}, checkpoint is restored!'.format(int(ckpt.step)))
return model
接着就是遍历列表,并读取图片数据,
# 遍历图片列表
for filename in image_filenames:
image = cv2.imread(filename)
print("filename:", filename)
# 根据文件名规则判断文件名中是否包含车牌信息,如果包含的话,解析出车牌号,用于跟模型预测的车牌号对比
if len(filename.split("-")) == 7:
_,number = get_plate_attribute(filename, image)
else:
number = "None"
然后,将图片缩放至宽和高都是32的整数倍,并且返回缩放比例,
'''
缩放图片至宽和高都是32的整数倍,且限定最大长度
'''
def east_resize_image(image, max_side_len=2400):
h, w, _ = image.shape
resize_w = w
resize_h = h
# limit the max side
if max(resize_h, resize_w) > max_side_len:
ratio = float(max_side_len) / resize_h if resize_h > resize_w else float(max_side_len) / resize_w
else:
ratio = 1.
resize_h = int(resize_h * ratio)
resize_w = int(resize_w * ratio)
resize_h = resize_h if resize_h % 32 == 0 else (resize_h // 32) * 32
resize_w = resize_w if resize_w % 32 == 0 else (resize_w // 32) * 32
image = cv2.resize(image, (int(resize_w), int(resize_h)))
ratio_h = resize_h / float(h)
ratio_w = resize_w / float(w)
return image, (ratio_h, ratio_w)
接着,就是预测出score map和geometry map了,
# 得到score_map和geo_map
score_map, geo_map = east_model.predict(image_resized[np.newaxis, :, :, :])
得到 score map和geometry map以后,先还原出所有score map预测出的文本框里所有像素点的预测出的文本框,代码如下,
'''
通过geometry map和origin得到文本框
思路:先根据旋转角度是的正负,以及像素点相对最小外接矩形的距离,在原点处还原出矩形。
再根据旋转角度得到旋转矩阵,然后求出5个关键点(四个顶点加像素点)旋转以后的坐标,
最后再根据旋转后的像素点坐标和实际的像素点坐标的相对位置,将旋转后的矩形平移,即可得到文本框坐标。
'''
def get_polys_by_geometry(origin, geometry, is_positive_angle):
# 得到距离
d = geometry[:, :4]
# 得到角度
angle = geometry[:, 4]
if is_positive_angle:
# 角度大于0的情况
origin = origin[angle >= 0]
d = d[angle >= 0]
angle = angle[angle >= 0]
else:
# 角度小于0的情况
origin = origin[angle < 0]
d = d[angle < 0]
angle = angle[angle < 0]
if origin.shape[0] > 0:
# top-0, right-1, bottom-2, left-3
d_top = d[:,0]
d_right = d[:,1]
d_bottom = d[:,2]
d_left = d[:,3]
d_w = d_left + d_right
d_h = d_top + d_bottom
p_zeros = np.zeros(d_top.shape[0])
if is_positive_angle:
# 角度大于0,则以p3顶点为原点,得到平行x轴的矩形
p0 = [p_zeros, -d_h]
p1 = [d_w, -d_h]
p2 = [d_w, p_zeros]
p3 = [p_zeros, p_zeros]
p_orgin = [d_left, -d_bottom] # 这个是像素点相对矩形的位置
# 旋转矩阵
rotate_matrix_x = np.array([np.cos(angle), np.sin(angle)]).transpose((1, 0))
rotate_matrix_y = np.array([-np.sin(angle), np.cos(angle)]).transpose((1, 0))
else:
# 角度大于0,则以p2顶点为原点,得到平行x轴的矩形
p0 = [-d_w, -d_h]
p1 = [p_zeros, -d_h]
p2 = [p_zeros, p_zeros]
p3 = [-d_w, p_zeros]
p_orgin = [-d_right, -d_bottom] # 这个是像素点相对矩形的位置
# 旋转矩阵
rotate_matrix_x = np.array([np.cos(-angle), -np.sin(-angle)]).transpose((1, 0))
rotate_matrix_y = np.array([np.sin(-angle), np.cos(-angle)]).transpose((1, 0))
# 旋转矩阵,因为总共要旋转5个点,所以要repeat出5个旋转矩阵
rotate_matrix_x = np.repeat(rotate_matrix_x, 5, axis=1).reshape(-1, 2, 5).transpose((0, 2, 1)) # N*5*2
rotate_matrix_y = np.repeat(rotate_matrix_y, 5, axis=1).reshape(-1, 2, 5).transpose((0, 2, 1))
# 将上面得到的5个点经过旋转矩阵得到旋转以后的点的坐标
p = np.asarray(np.concatenate([p0, p1, p2, p3, p_orgin])).transpose((1, 0)).reshape((-1, 5, 2)) # N*5*2
p_rotate_x = np.sum(rotate_matrix_x * p, axis=2)[:, :, np.newaxis] # N*5*1
p_rotate_y = np.sum(rotate_matrix_y * p, axis=2)[:, :, np.newaxis] # N*5*1
# 这里得到的是旋转后的坐标
p_rotate = np.concatenate([p_rotate_x, p_rotate_y], axis=2) # N*5*2
# 根据旋转后的像素点坐标和原像素点坐标的相对位置,平移整个矩形
p3_in_origin = origin - p_rotate[:, 4, :]
new_p0 = p_rotate[:, 0, :] + p3_in_origin
new_p1 = p_rotate[:, 1, :] + p3_in_origin
new_p2 = p_rotate[:, 2, :] + p3_in_origin
new_p3 = p_rotate[:, 3, :] + p3_in_origin
# 得到最终的文本框
new_p_0 = np.concatenate([new_p0[:, np.newaxis, :], new_p1[:, np.newaxis, :],
new_p2[:, np.newaxis, :], new_p3[:, np.newaxis, :]], axis=1) # N*5*2
else:
new_p_0 = np.zeros((0, 4, 2))
return new_p_0
'''
根据score map和geometry map得到所有相关像素点的文本框
'''
def restore_rectangle(origin, geometry):
return np.concatenate([get_polys_by_geometry(origin, geometry, is_positive_angle=True),
get_polys_by_geometry(origin, geometry, is_positive_angle=False)])
得到这些文本框以后,再通过lanms得到最终的文本框,
'''
east预测车牌坐标
'''
def east_detect(score_map, geo_map, timer, score_map_thresh=0.8, box_thresh=0.1, nms_thres=0.2):
np.set_printoptions(threshold=np.inf)
if len(score_map.shape) == 4:
score_map = score_map[0, :, :, 0]
geo_map = geo_map[0, :, :, ]
# filter the score map
# 过滤得分小于score_map_thresh的区域
xy_text = np.argwhere(score_map > score_map_thresh)
# sort the text boxes via the y axis
xy_text = xy_text[np.argsort(xy_text[:, 0])]
# restore
start = time.time()
text_box_restored = restore_rectangle(xy_text[:, ::-1]*4, geo_map[xy_text[:, 0], xy_text[:, 1], :]) # N*4*2
# print('{} text boxes before nms'.format(text_box_restored.shape[0]))
boxes = np.zeros((text_box_restored.shape[0], 9), dtype=np.float32)
boxes[:, :8] = text_box_restored.reshape((-1, 8))
boxes[:, 8] = score_map[xy_text[:, 0], xy_text[:, 1]]
timer['restore'] = time.time() - start
# nms part
start = time.time()
# boxes = nms_locality.nms_locality(boxes.astype(np.float64), nms_thres)
boxes = lanms.merge_quadrangle_n9(boxes.astype('float32'), nms_thres)
timer['nms'] = time.time() - start
if boxes.shape[0] == 0:
return None, timer
# here we filter some low score boxes by the average score map, this is different from the orginal paper
for i, box in enumerate(boxes):
mask = np.zeros_like(score_map, dtype=np.uint8)
cv2.fillPoly(mask, box[:8].reshape((-1, 4, 2)).astype(np.int32) // 4, 1)
boxes[i, 8] = cv2.mean(score_map, mask)[0]
boxes = boxes[boxes[:, 8] > box_thresh]
return boxes, timer
得到文本框坐标以后,就要把车牌单独截取出来了,先遍历所有文本框,过滤太小的文本框,
# 如果有车牌坐标,则去识别车牌号
if boxes is not None:
# 先将车牌坐标恢复成跟缩放前图片对应的坐标
boxes = boxes[:, :8].reshape((-1, 4, 2))
boxes[:, :, 0] /= ratio_w
boxes[:, :, 1] /= ratio_h
# 遍历所有识别出来的车牌坐标
for box in boxes:
# 重新排序顶点
box = east_sort_poly(box.astype(np.int32))
# 过滤太小的坐标或者小于0的坐标
if np.linalg.norm(box[0] - box[1]) < 10 or np.linalg.norm(box[3]-box[0]) < 10 or np.min(box) < 0:
continue
然后,获取文本框的最小外接矩形,再重新排序顶点,接着旋转图片以使最小外接矩形水平,然后就可以截取出车牌了。代码如下,
def get_min_area_rect(poly):
rect = cv2.minAreaRect(poly)
box = cv2.boxPoints(rect)
return box, rect[2]
'''
对矩形的顶点进行排序,排序后的结果是,p0-左上角,p1-右上角,p2-右下角,p3-左下角
矩形与水平轴的夹角,即为p2_p3边到x轴的夹角,以逆时针为正,顺时针为负
'''
def sort_poly_and_get_angle(poly, image=None):
# 先找到矩形最底部的顶点
p_lowest = np.argmax(poly[:, 1])
if np.count_nonzero(poly[:, 1] == poly[p_lowest, 1]) == 2:
# 如果矩形底部的边刚好与x轴平行
# 这种情况下,x坐标加y坐标之和最小的顶点就是左上角的顶点,即p0
p0_index = np.argmin(np.sum(poly, axis=1))
p1_index = (p0_index + 1) % 4
p2_index = (p0_index + 2) % 4
p3_index = (p0_index + 3) % 4
return poly[[p0_index, p1_index, p2_index, p3_index]], 0.
else:
# 如果矩形底部与x轴有夹角
# 找到最底部顶点的右边的第一个顶点
p_lowest_right = (p_lowest - 1) % 4
# 求最底部顶点与其右边第一个顶点组成的边与x轴的夹角
angle = np.arctan(-(poly[p_lowest][1] - poly[p_lowest_right][1])/(poly[p_lowest][0] - poly[p_lowest_right][0] + 1e-5))
# 下面的代码其实自己画个图就很好理解了
if angle > np.pi/4:
# 如果这个夹角大于45度,那么,最底部的顶点为p2顶点
p2_index = p_lowest
p1_index = (p2_index - 1) % 4
p0_index = (p2_index - 2) % 4
p3_index = (p2_index + 1) % 4
# 这种情况下,p2-p3边与x轴的夹角就为-(np.pi/2 - angle)
return poly[[p0_index, p1_index, p2_index, p3_index]], -(np.pi/2 - angle)
else:
# 如果这个夹角小于等于45度,那么,最底部的顶点为p3顶点
p3_index = p_lowest
p0_index = (p3_index + 1) % 4
p1_index = (p3_index + 2) % 4
p2_index = (p3_index + 3) % 4
return poly[[p0_index, p1_index, p2_index, p3_index]], angle
def get_plate_image(image, poly):
# DEBUG = True
if DEBUG:
image = draw_line(image, poly)
box, angle = get_min_area_rect(poly)
if DEBUG:
image = draw_line(image, box, (0,255,0))
(p0, p1, p2, p3), angle = sort_poly_and_get_angle(box)
image = crop_plate(image, angle, p0, p1, p2, p3)
# cv2.imshow("get_min_area_rect", image)
# cv2.waitKey(0)
return image
# 将车牌裁剪出来,因为车牌有可能是斜的,所以要先将图片旋转到车牌的矩形框为水平时,再裁剪
def crop_plate(image, angle, p0, p1, p2, p3):
# DEBUG = True
angle = -angle * (180 / math.pi)
# print("angle:", angle)
h, w, _ = image.shape
rotate_mat = cv2.getRotationMatrix2D((w / 2, h / 2), angle, 1) # 按angle角度旋转图像
h_new = int(w * np.fabs(np.sin(np.radians(angle))) + h * np.fabs(np.cos(np.radians(angle))))
w_new = int(h * np.fabs(np.sin(np.radians(angle))) + w * np.fabs(np.cos(np.radians(angle))))
rotate_mat[0, 2] += (w_new - w) / 2
rotate_mat[1, 2] += (h_new - h) / 2
rotated_image = cv2.warpAffine(image, rotate_mat, (w_new, h_new), borderValue=(255, 255, 255))
# 旋转后图像的四点坐标
[[p1[0]], [p1[1]]] = np.dot(rotate_mat, np.array([[p1[0]], [p1[1]], [1]]))
[[p3[0]], [p3[1]]] = np.dot(rotate_mat, np.array([[p3[0]], [p3[1]], [1]]))
[[p2[0]], [p2[1]]] = np.dot(rotate_mat, np.array([[p2[0]], [p2[1]], [1]]))
[[p0[0]], [p0[1]]] = np.dot(rotate_mat, np.array([[p0[0]], [p0[1]], [1]]))
if DEBUG:
cv2.circle(rotated_image, tuple(p0), 10, (0,255,0), 4)
cv2.circle(rotated_image, tuple(p1), 10, (0,0,255), 4)
cv2.imshow('image', image)
cv2.imshow('rotateImg2', rotated_image)
cv2.waitKey(0)
crop_image = rotated_image[int(p0[1]):int(p3[1]), int(p0[0]):int(p1[0])]
return crop_image
3、车牌号识别
得到车牌以后,就要对车牌进行字符识别了,先导入DenseNet模型,
'''
创建和加载densenet模型
'''
def densenet_create_and_load_model():
_,char_list = get_char_vector(FLAGS.char_filename)
num_classes = len(char_list) + 1
model,_,_ = densenet(FLAGS, num_classes)
h5_path = os.path.join("./densenet/models", "plate-"+FLAGS.densenet_model_net+".h5")
if os.path.exists(h5_path):
model.load_weights(h5_path)
return model, char_list, num_classes
然后将车牌图片缩放至固定高度,代码如下,
def pading_plate_width(image):
h, w, _ = image.shape
new_w = np.where(FLAGS.input_size_w > w+20, FLAGS.input_size_w, w+20)
new_image = np.zeros((h, new_w, 3), dtype=np.uint8)
start_w = 10
new_image[:h,start_w:w+start_w,:] = image[:h,:w,:]
return new_image
'''
随机旋转,这里不旋转一下识别效果反而降低,可能是训练的时候大部分都旋转的原因
'''
def random_rotate(images):
h, w, _ = images.shape
random_angle = np.random.randint(-15,15)
random_scale = np.random.randint(8,10) / 10.0
mat_rotate = cv2.getRotationMatrix2D(center=(w*0.5, h*0.5), angle=random_angle, scale=random_scale)
images = cv2.warpAffine(images, mat_rotate, (w, h))
return images
'''
将图片缩放至固定高度,且保持原来的宽高比
'''
def densenet_resize_image(image):
image = random_rotate(image)
h, w, _ = image.shape
new_w = np.around(w / (h/FLAGS.input_size_h)).astype(np.int32)
image = cv2.resize(image, (new_w, FLAGS.input_size_h))
image = pading_plate_width(image)
# cv2.imshow("image", image)
image = image[np.newaxis,:,:,:]
return image
接着直接将图片输入到模型中,得到预测结果,
# densenet模型识别车牌号
y_pred = densenet_model.predict(resized_plate_image)
对上面的预测结果还不能直接用,还要先删除掉ctc的blank字符和重复字符,再将其转成文字,
'''
解析densenet的预测结果,将blank和重复字符去掉
'''
def densenet_decode(pred, char_list, num_classes):
plate_char_list = []
pred_text = pred.argmax(axis=2)[0]
# print("pred_text:", pred_text, " char_list len:", len(char_list))
for i in range(len(pred_text)):
if pred_text[i] != num_classes - 1 and ((not (i > 0 and pred_text[i] == pred_text[i - 1])) or (i > 1 and pred_text[i] == pred_text[i - 2])):
plate_char_list.append(char_list[pred_text[i]])
return u''.join(plate_char_list)
这样,我们就得到了车牌号了,为了方便查看,将识别结果显示出来,
'''
显示预测的车牌号
'''
def show_plate_number(image, number, coor):
h, w, _ = image.shape
font = ImageFont.truetype(FLAGS.fontpath, 25)
draw = ImageDraw.Draw(Image.fromarray(image))
textsize = draw.textsize(number, font=font)
end_x = coor[0] + textsize[0]
end_x = np.where(end_x > w, w, end_x)
image[coor[1]:coor[1]+textsize[1], coor[0]:end_x] = 0
img = Image.fromarray(image)
draw = ImageDraw.Draw(img)
draw.text(coor, number, font = font, fill = (0,255,255,0))
image = np.array(img)
return image
4、运行结果
现在来看运行结果,
效果还是可以的。