本文主要针对图像金字塔进行模版匹配，可以匹配到一幅图像中多个目标。
一、主要流程：
- 主要对模板构建图像金字塔
- 利用opencv对1中得到模板图像，进行模版匹配
- 对2中得到的匹配框进行nms处理
- 对3中符合要求匹配框进行nms处理，最后的检测框即为符合的匹配框
二、nms处理代码
#!/usr/bin/env python
# -*- encoding: utf-8 -*-
'''
Descripttion:
version: 1.0
Author:
Date: 2020-09-21 17:29:05
LastEditors:
LastEditTime: 2020-09-29 16:55:43
Software:: Visual Studio Code
FilePath: nms.py
'''
import numpy as np
def py_cpu_nms(dets, thresh):
x1 = dets[:, 0]
y1 = dets[:, 1]
x2 = dets[:, 2]
y2 = dets[:, 3]
scores = dets[:, 4]
areas = (x2 - x1 + 1) * (y2 - y1 + 1)
order = scores.argsort()[::-1]
keep = []
while order.size > 0:
i = order[0]
keep.append(i)
xx1 = np.maximum(x1[i], x1[order[1:]])
yy1 = np.maximum(y1[i], y1[order[1:]])
xx2 = np.minimum(x2[i], x2[order[1:]])
yy2 = np.minimum(y2[i], y2[order[1:]])
w = np.maximum(0.0, xx2 - xx1 + 1)
h = np.maximum(0.0, yy2 - yy1 + 1)
inter = w * h
ovr = inter / (areas[i] + areas[order[1:]] - inter)
inds = np.where(ovr <= thresh)[0]
order = order[inds + 1]
return keep
三、模板匹配代码
#!/usr/bin/env python
# -*- encoding: utf-8 -*-
'''
Descripttion:
version: 1.0
Author:
Date: 2020-09-21 17:29:05
LastEditors:
LastEditTime: 2020-09-29 16:55:43
Software:: Visual Studio Code
FilePath: image_match.py
'''
from cv2 import cv2
import numpy as np
from nms import py_cpu_nms
import time
# 金字塔缩放参数，需要根据实际情况进行调整
scale_factor = 1 / 1.2
# 图像缩放
def processed_image(image, scale):
height, width = image.shape
new_height = int(height * scale)
new_width = int(width * scale)
new_dim = (new_width, new_height)
img_resized = cv2.resize(image, new_dim, interpolation=cv2.INTER_LINEAR)
return img_resized
# 图像金字塔
def pyramid_image(image, current_scale=1.0):
min_h = 12
min_w = 12
dim_h = 0
dim_w = 0
ratios = []
images = []
if len(image.shape) < 3:
dim_h, dim_w = image.shape
else:
dim_h, dim_w, dim_c = image.shape
while (dim_h > min_h and dim_w > min_w):
current_scale *= scale_factor
resized_img = processed_image(image, current_scale)
images.append(resized_img)
ratios.append(current_scale)
dim_h, dim_w = resized_img.shape
return images, ratios
def image_match(gray_src, tpl, methods, threshold=0.62):
box_rec = []
h, w = tpl.shape[:2]
# 模板不能大于图像
if h > min(gray_src.shape[:2]) or w > min(gray_src.shape[:2]):
return box_rec
res = cv2.matchTemplate(gray_src, tpl, methods)
min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(res)
print(max_val)
loc = np.where(res >= threshold)
for pt in zip(*loc[::-1]):
xmin = pt[0]
ymin = pt[1]
xmax = pt[0] + w
ymax = pt[1] + h
score = res[pt[1]][pt[0]]
# 搜索框在图像左边，排除右边的干扰区域
if xmax < gray_src.shape[1] / 2:
box_rec.append([xmin, ymin, xmax, ymax, score])
box_rec = np.array(box_rec)
if len(box_rec) != 0:
keep = py_cpu_nms(box_rec, 0.3)
box_rec = box_rec[keep]
return box_rec
def pyramid_image_match(image,template,threshold=0.62):
result_box = []
if image is None or template_white is None or template_black is None:
return np.array(result_box)
h_ratio = 180. / 1080.
height, width, channel = image.shape
roi_y =0
roi_h = int(h_ratio * height)
roi_img = image[roi_y:roi_y + roi_h, 0:width]
gray_roi_img = cv2.cvtColor(roi_img, cv2.COLOR_BGR2GRAY)
gray_tem = cv2.cvtColor(template, cv2.COLOR_BGR2GRAY)
current_scale = max(1.0, max(height * 1.0 / width, height / 1080.0))
methods = cv2.TM_CCOEFF_NORMED
if current_scale != 2339 / 1080.0:
tmp_py_list, ratio_py_list = pyramid_image(gray_tem, current_scale)
for tpl, tpl_ratio in zip(tmp_py_list, ratio_py_list):
box_rec = image_match(gray_roi_img, tpl, methods, threshold)
if len(box_rec) != 0:
result_box.append(box_rec)
else:
box_rec = image_match(gray_roi_img, gray_tem, methods, threshold)
if len(box_rec) != 0:
result_box.append(box_rec)
result_box = np.array(result_box)
if result_box.size != 0:
result_box = np.vstack(result_box)
keep = py_cpu_nms(result_box, 0.3)
result_box = result_box[keep]
result_box.astype(int)
return result_box
if __name__ == '__main__':
image = cv2.imread("image.jpg")
template = cv2.imread("template.jpg")
start = time.time()
box_res = pyramid_image_match(image, template)
if box_res.size != 0:
cv2.namedWindow('image', cv2.WINDOW_NORMAL)
for box in box_res:
box = box[:4].astype(int)
cv2.rectangle(image, (box[0], box[1]), (box[2], box[3]),
(0, 0, 255), 2)
cv2.imshow("image", image)
cv2.waitKey(0)
