devbox
花生_TL007
这个屌丝很懒，什么也没留下！
关注作者
热门标签
热门文章
当前位置:   article > 正文

【Yolov8 Opencv C++系列保姆教程】Yolov8 opencv c++ 版本保姆教程，Yolov8训练自己的数据集，实现红绿灯识别及红绿灯故障检测 ，红绿灯故障识别。_yolo红绿灯检测

作者：花生_TL007 | 2024-05-18 10:52:10

赞

踩

yolo红绿灯检测

目录

一、Yolov8简介

1、yolov8 源码地址：

2、官方文档：

3、预训练模型百度网盘地址：

二、模型训练

1、标定红绿灯数据：

2、训练环境：

3、数据转化：

4、构造训练数据：

5、训练样本：

三、验证模型：

1、图像测试：

2、视频测试：

四、导出ONNX

五、Opencv实现Yolov8 C++ 识别

1、开发环境：

2、main函数代码：

3、yolov8 头文件inference.h代码：

4、yolov8 cpp文件inference.cpp代码：

一、Yolov8简介

1、yolov8 源码地址：

工程链接：https://github.com/ultralytics/ultralytics

2、官方文档：

CLI - Ultralytics YOLOv8 Docs

3、预训练模型百度网盘地址：

训练时需要用到，下载的网址较慢：

如果模型下载不了，加QQ：187100248.

链接: https://pan.baidu.com/s/1YfMxRPGk8LF75a4cbgYxGg 提取码: rd7b

二、模型训练

1、标定红绿灯数据：

         类别为23类，分别为：

红绿灯类别
red_lightgreen_lightyellow_lightoff_lightpart_ry_lightpart_rg_light
part_yg_lightryg_lightcountdown_off_lightcountdown_on_lightshade_lightzero
onetwothreefourfivesix
seveneightninebrokeNumberbrokenLight

        标注工具地址：AI标注工具Labelme和LabelImage Labelme和LabelImage集成工具_labelimage与labelme-CSDN博客

标注后图像格式

2、训练环境：

1）、Ubuntu18.04；

2）、Cuda11.7 + CUDNN8.0.6；

3）、opencv4.5.5；

4）、PyTorch1.8.1-GPU；

5）、python3.9

3、数据转化：

 1）、需要把上面标定的数据集中的.xml文件转换为.txt，转换代码为：

  1. import os
  2. import shutil
  3. import xml.etree.ElementTree as ET
  4. from xml.etree.ElementTree import Element, SubElement
  5. from PIL import Image
  6. import cv2
  7.  
  8. classes = ['red_light', 'green_light', 'yellow_light', 'off_light', 'part_ry_light', 'part_rg_light', 'part_yg_light', 'ryg_light',
  9.            'countdown_off_light', 'countdown_on_light','shade_light','zero','one','two','three','four','five','six','seven',
  10.            'eight','nine','brokeNumber','brokenLight']
  11.  
  12. class Xml_make(object):
  13.     def __init__(self):
  14.         super().__init__()
  15.  
  16.     def __indent(self, elem, level=0):
  17.         i = "\n" + level * "\t"
  18.         if len(elem):
  19.             if not elem.text or not elem.text.strip():
  20.                 elem.text = i + "\t"
  21.             if not elem.tail or not elem.tail.strip():
  22.                 elem.tail = i
  23.             for elem in elem:
  24.                 self.__indent(elem, level + 1)
  25.             if not elem.tail or not elem.tail.strip():
  26.                 elem.tail = i
  27.         else:
  28.             if level and (not elem.tail or not elem.tail.strip()):
  29.                 elem.tail = i
  30.  
  31.     def _imageinfo(self, list_top):
  32.         annotation_root = ET.Element('annotation')
  33.         annotation_root.set('verified', 'no')
  34.         tree = ET.ElementTree(annotation_root)
  35.         '''
  36.         0:xml_savepath 1:folder,2:filename,3:path
  37.         4:checked,5:width,6:height,7:depth
  38.         '''
  39.         folder_element = ET.Element('folder')
  40.         folder_element.text = list_top[1]
  41.         annotation_root.append(folder_element)
  42.  
  43.         filename_element = ET.Element('filename')
  44.         filename_element.text = list_top[2]
  45.         annotation_root.append(filename_element)
  46.  
  47.         path_element = ET.Element('path')
  48.         path_element.text = list_top[3]
  49.         annotation_root.append(path_element)
  50.  
  51.         # checked_element = ET.Element('checked')
  52.         # checked_element.text = list_top[4]
  53.         # annotation_root.append(checked_element)
  54.  
  55.         source_element = ET.Element('source')
  56.         database_element = SubElement(source_element, 'database')
  57.         database_element.text = 'Unknown'
  58.         annotation_root.append(source_element)
  59.  
  60.         size_element = ET.Element('size')
  61.         width_element = SubElement(size_element, 'width')
  62.         width_element.text = str(list_top[5])
  63.         height_element = SubElement(size_element, 'height')
  64.         height_element.text = str(list_top[6])
  65.         depth_element = SubElement(size_element, 'depth')
  66.         depth_element.text = str(list_top[7])
  67.         annotation_root.append(size_element)
  68.  
  69.         segmented_person_element = ET.Element('segmented')
  70.         segmented_person_element.text = '0'
  71.         annotation_root.append(segmented_person_element)
  72.  
  73.         return tree, annotation_root
  74.  
  75.     def _bndbox(self, annotation_root, list_bndbox):
  76.         for i in range(0, len(list_bndbox), 9):
  77.             object_element = ET.Element('object')
  78.             name_element = SubElement(object_element, 'name')
  79.             name_element.text = list_bndbox[i]
  80.  
  81.             # flag_element = SubElement(object_element, 'flag')
  82.             # flag_element.text = list_bndbox[i + 1]
  83.  
  84.             pose_element = SubElement(object_element, 'pose')
  85.             pose_element.text = list_bndbox[i + 2]
  86.  
  87.             truncated_element = SubElement(object_element, 'truncated')
  88.             truncated_element.text = list_bndbox[i + 3]
  89.  
  90.             difficult_element = SubElement(object_element, 'difficult')
  91.             difficult_element.text = list_bndbox[i + 4]
  92.  
  93.             bndbox_element = SubElement(object_element, 'bndbox')
  94.             xmin_element = SubElement(bndbox_element, 'xmin')
  95.             xmin_element.text = str(list_bndbox[i + 5])
  96.  
  97.             ymin_element = SubElement(bndbox_element, 'ymin')
  98.             ymin_element.text = str(list_bndbox[i + 6])
  99.  
  100.             xmax_element = SubElement(bndbox_element, 'xmax')
  101.             xmax_element.text = str(list_bndbox[i + 7])
  102.  
  103.             ymax_element = SubElement(bndbox_element, 'ymax')
  104.             ymax_element.text = str(list_bndbox[i + 8])
  105.  
  106.             annotation_root.append(object_element)
  107.  
  108.         return annotation_root
  109.  
  110.     def txt_to_xml(self, list_top, list_bndbox):
  111.         tree, annotation_root = self._imageinfo(list_top)
  112.         annotation_root = self._bndbox(annotation_root, list_bndbox)
  113.         self.__indent(annotation_root)
  114.         tree.write(list_top[0], encoding='utf-8', xml_declaration=True)
  115.  
  116. def txt_2_xml(source_path, xml_save_dir, jpg_save_dir,txt_dir):
  117.  
  118.     COUNT = 0
  119.     for folder_path_tuple, folder_name_list, file_name_list in os.walk(source_path):
  120.         for file_name in file_name_list:
  121.             file_suffix = os.path.splitext(file_name)[-1]
  122.             if file_suffix != '.jpg':
  123.                 continue
  124.             list_top = []
  125.             list_bndbox = []
  126.             path = os.path.join(folder_path_tuple, file_name)
  127.             xml_save_path = os.path.join(xml_save_dir, file_name.replace(file_suffix, '.xml'))
  128.             txt_path = os.path.join(txt_dir, file_name.replace(file_suffix, '.txt'))
  129.             filename = file_name#os.path.splitext(file_name)[0]
  130.             checked = 'NO'
  131.             #print(file_name)
  132.             im = Image.open(path)
  133.             im_w = im.size[0]
  134.             im_h = im.size[1]
  135.  
  136.             shutil.copy(path, jpg_save_dir)
  137.  
  138.  
  139.             if im_w*im_h > 34434015:
  140.                 print(file_name)
  141.             if im_w < 100:
  142.                 print(file_name)
  143.  
  144.             width = str(im_w)
  145.             height = str(im_h)
  146.             depth = '3'
  147.             flag = 'rectangle'
  148.             pose = 'Unspecified'
  149.             truncated = '0'
  150.             difficult = '0'
  151.             list_top.extend([xml_save_path, folder_path_tuple, filename, path, checked, width, height, depth])
  152.             for line in open(txt_path, 'r'):
  153.                 line = line.strip()
  154.                 info = line.split(' ')
  155.                 name = classes[int(info[0])]
  156.                 x_cen = float(info[1]) * im_w
  157.                 y_cen = float(info[2]) * im_h
  158.                 w = float(info[3]) * im_w
  159.                 h = float(info[4]) * im_h
  160.                 xmin = int(x_cen - w / 2) - 1
  161.                 ymin = int(y_cen - h / 2) - 1
  162.                 xmax = int(x_cen + w / 2) + 3
  163.                 ymax = int(y_cen + h / 2) + 3
  164.  
  165.                 if xmin < 0:
  166.                     xmin = 0
  167.                 if ymin < 0:
  168.                     ymin = 0
  169.  
  170.                 if xmax > im_w - 1:
  171.                     xmax = im_w - 1
  172.                 if ymax > im_h - 1:
  173.                     ymax = im_h - 1
  174.  
  175.                 if w > 5 and h > 5:
  176.                   list_bndbox.extend([name, flag, pose, truncated, difficult,str(xmin), str(ymin), str(xmax), str(ymax)])
  177.  
  178.                 if xmin < 0 or xmax > im_w - 1 or ymin < 0 or ymax > im_h - 1:
  179.                     print(xml_save_path)
  180.  
  181.             Xml_make().txt_to_xml(list_top, list_bndbox)
  182.             COUNT += 1
  183.             #print(COUNT, xml_save_path)
  184.  
  185. if __name__ == "__main__":
  186.  
  187.     out_xml_path = "/home/TL_TrainData/"  # .xml输出文件存放地址
  188.     out_jpg_path = "/home/TL_TrainData/"  # .jpg输出文件存放地址
  189.     txt_path = "/home/Data/TrafficLight/trainData"  # yolov3标注.txt和图片文件夹
  190.     images_path = "/home/TrafficLight/trainData"  # image文件存放地址
  191.     txt_2_xml(images_path, out_xml_path, out_jpg_path, txt_path)

4、构造训练数据：

2）、训练样本数据构造，需要把分成images和labels，images下面放入图片，labels下面放入.txt文件:

分成images和labels
分成images和labels
images
labels

5、训练样本：

 1）、首先安装训练包：

pip install ultralytics

2）、修改训练数据参数coco128_light.yaml文件，这个是自己修改的。

  1. # Ultralytics YOLO 
    声明：本文内容由网友自发贡献，不代表【wpsshop博客】立场，版权归原作者所有，本站不承担相应法律责任。如您发现有侵权的内容，请联系我们。转载请注明出处：https://www.wpsshop.cn/w/花生_TL007/article/detail/587951
    推荐阅读
    相关标签

Copyright © 2003-2013 www.wpsshop.cn 版权所有，并保留所有权利。