使用YOLOv2进行图像检测

时间:2024-01-23 08:13:17
  • 基本配置信息

tensorflow (1.4.0)

tensorflow-tensorboard (0.4.0)

Keras (2.1.5)

Python (3.6.0)

Anaconda 4.3.1 (64-bit)

Windows 7

  • darknet链接

https://github.com/pjreddie/darknet

下载后在cfg文件夹下找到yolov2的配置文件yolov2.cfg

 

  • yolov2权重文件链接

https://pjreddie.com/darknet/yolov2/

在页面中选择YOLOV2 weights下载

 

  • yad2k 链接

https://github.com/allanzelener/YAD2K

下载完成后将之前下载好的yolov2.cfg文件,YOLOV2 weights文件拷贝到yad2k目录下

 

 

  • 使用spyder 运行yad2k目录下的yad2k.py文件

在运行配置里设置运行时所需的参数信息

 

或使用命令行运行yad2k.py

python yad2k.py yolov2.cfg yolov2.weights model_data/yolo.h5

运行结果如图所示

 

生成的yolo.h5文件在model_data文件夹内

 

 

  • 利用生成的权重信息,进行图像检测

使用opencv调用电脑摄像头,进行视频图像信息的检测

opencv版本

opencv-python (3.2.0)

yad2k目录下创建自己的demo,参考https://www.jianshu.com/p/3e77cefeb49b

 

  1 import cv2
  2 import os
  3 import time
  4 import numpy as np
  5 from keras import backend as K
  6 from keras.models import load_model
  7 
  8 from yad2k.models.keras_yolo import yolo_eval, yolo_head
  9 
 10 
 11 class YOLO(object):
 12     def __init__(self):
 13         self.model_path = \'model_data/yolo.h5\'
 14         self.anchors_path = \'model_data/yolo_anchors.txt\'
 15         self.classes_path = \'model_data/coco_classes.txt\'
 16         self.score = 0.3
 17         self.iou = 0.5
 18 
 19         self.class_names = self._get_class()
 20         self.anchors = self._get_anchors()
 21         self.sess = K.get_session()
 22         self.boxes, self.scores, self.classes = self.generate()
 23 
 24     def _get_class(self):
 25         classes_path = os.path.expanduser(self.classes_path)
 26         with open(classes_path) as f:
 27             class_names = f.readlines()
 28         class_names = [c.strip() for c in class_names]
 29         return class_names
 30 
 31     def _get_anchors(self):
 32         anchors_path = os.path.expanduser(self.anchors_path)
 33         with open(anchors_path) as f:
 34             anchors = f.readline()
 35             anchors = [float(x) for x in anchors.split(\',\')]
 36             anchors = np.array(anchors).reshape(-1, 2)
 37         return anchors
 38 
 39     def generate(self):
 40         model_path = os.path.expanduser(self.model_path)
 41         assert model_path.endswith(\'.h5\'), \'Keras model must be a .h5 file.\'
 42 
 43         self.yolo_model = load_model(model_path)
 44 
 45         # Verify model, anchors, and classes are compatible
 46         num_classes = len(self.class_names)
 47         num_anchors = len(self.anchors)
 48         # TODO: Assumes dim ordering is channel last
 49         model_output_channels = self.yolo_model.layers[-1].output_shape[-1]
 50         assert model_output_channels == num_anchors * (num_classes + 5), \
 51             \'Mismatch between model and given anchor and class sizes\'
 52         print(\'{} model, anchors, and classes loaded.\'.format(model_path))
 53 
 54         # Check if model is fully convolutional, assuming channel last order.
 55         self.model_image_size = self.yolo_model.layers[0].input_shape[1:3]
 56         self.is_fixed_size = self.model_image_size != (None, None)
 57 
 58         # Generate output tensor targets for filtered bounding boxes.
 59         # TODO: Wrap these backend operations with Keras layers.
 60         yolo_outputs = yolo_head(self.yolo_model.output, self.anchors, len(self.class_names))
 61         self.input_image_shape = K.placeholder(shape=(2, ))
 62         boxes, scores, classes = yolo_eval(yolo_outputs, self.input_image_shape, score_threshold=self.score, iou_threshold=self.iou)
 63         return boxes, scores, classes
 64 
 65     def detect_image(self, image):
 66         start = time.time()
 67         #image = cv2.imread(image)
 68         #cv2.imshow(\'image\',image)  
 69         y, x, _ = image.shape
 70 
 71         if self.is_fixed_size:  # TODO: When resizing we can use minibatch input.
 72             resized_image = cv2.resize(image, tuple(reversed(self.model_image_size)), interpolation=cv2.INTER_CUBIC)
 73             image_data = np.array(resized_image, dtype=\'float32\')
 74         else:
 75             image_data = np.array(image, dtype=\'float32\')
 76 
 77         image_data /= 255.
 78         image_data = np.expand_dims(image_data, 0)  # Add batch dimension.
 79 
 80         out_boxes, out_scores, out_classes = self.sess.run(
 81             [self.boxes, self.scores, self.classes],
 82             feed_dict={
 83                 self.yolo_model.input: image_data,
 84                 self.input_image_shape: [image.shape[0], image.shape[1]],
 85                 K.learning_phase(): 0
 86             })
 87         print(\'Found {} boxes for {}\'.format(len(out_boxes), \'img\'))
 88 
 89         for i, c in reversed(list(enumerate(out_classes))):
 90             predicted_class = self.class_names[c]
 91             box = out_boxes[i]
 92             score = out_scores[i]
 93 
 94             label = \'{} {:.2f}\'.format(predicted_class, score)
 95             top, left, bottom, right = box
 96             top = max(0, np.floor(top + 0.5).astype(\'int32\'))
 97             left = max(0, np.floor(left + 0.5).astype(\'int32\'))
 98             bottom = min(y, np.floor(bottom + 0.5).astype(\'int32\'))
 99             right = min(x, np.floor(right + 0.5).astype(\'int32\'))
100             print(label, (left, top), (right, bottom))
101 
102             cv2.rectangle(image, (left, top), (right, bottom), (255, 0, 0), 2)
103             cv2.putText(image, label, (left, int(top - 4)), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255), 1, cv2.LINE_AA)
104         end = time.time()
105         print(end - start)
106         return image
107 
108     def close_session(self):
109         self.sess.close()
110 
111  
112 def detect_vedio(yolo):
113     camera = cv2.VideoCapture(0)
114    
115     while True:
116         res, frame = camera.read()
117 
118         if not res:
119             break
120 
121         image = yolo.detect_image(frame)
122         cv2.imshow("detection", image)
123        
124         if cv2.waitKey(1) & 0xFF == ord(\'q\'):
125             break
126     yolo.close_session()
127 
128 
129 def detect_img(img, yolo):
130     image = cv2.imread(img)
131     r_image = yolo.detect_image(image)
132     cv2.namedWindow("detection")
133     while True:
134         cv2.imshow("detection", r_image)
135         if cv2.waitKey(110) & 0xff == 27:
136                 break
137     yolo.close_session()
138 
139 
140 if __name__ == \'__main__\':
141     yolo = YOLO()
142     detect_vedio(yolo)