Python在视频处理上的优势有哪些

妖狐艹你老母 2022-05-27 01:14 157阅读 0赞

今天与大家分享一些Python处理视频的一下经验，视频的处理和图片的处理类似，只不过视频处理需要连续处理一系列图片。

![20180409103820850][]

一般有两种视频源，一种是直接从硬盘加载视频，另一种是获取摄像头视频。

一共这几个模块：

class videoReader 读取视频

class videoFramesExtractor(videoReader):继承了读取视频，主要是用来限制读取视频中的哪些帧，并保存。

read\_excel\_single(excel\_path,event\_instance): 处理Excel中的数组，找到事件的帧数范围：

Excel内容是这样的，需要的只是第二列的帧范围，另外第二列中的这些值还有重复的所以需要去除重复：

每一个范围，例如:23450:23461之内的帧要存放在一个文件夹里。

![SouthEast][]

**\[python\]** [view plain][] [copy][view plain]

1.  </pre><pre name="code" **class="python">\# -\*- coding: cp936 -\*-** 
2.  
3.  **import cv2.cv as cv** 
4.  **import os** 
5.  **import cv2** 
6.  **import numpy** 
7.  **import Image** 
8.  **import xlrd** 
9.  
10. **class videoReader:** 
11. frame\_count=0 
12. **def \_\_init\_\_(self,videoPath):** 
13. self.videoPath=videoPath 
14. **def video\_init(self):** 
15. self.capture = cv.CaptureFromFile(self.videoPath) 
16. self.win\_name = "test" 
17. cv.NamedWindow(self.win\_name, cv.CV\_WINDOW\_AUTOSIZE) 
18. 
19. **def read(self):** 
20. self.video\_init() 
21. **while 1:** 
22. self.frame\_count+=1 
23. image = cv.QueryFrame(self.capture ) 
24. 
25. cv.ShowImage(self.win\_name, image) 
26. **print self.frame\_count** 
27. c = cv.WaitKey(10) 
28. **if c == 27:** 
29. **break** 
30. 
31. cv.DestroyWindow(self.win\_name) 
32. 
33. 
34. **class videoFramesExtractor(videoReader):** 
35. save\_path='d:/' 
36. **def \_\_init\_\_(self,videoPath,frameSpan,image\_prefix):\#image\_prefix=event\_name** 
37. videoReader.\_\_init\_\_(self,videoPath) 
38. self.frameSpan=frameSpan 
39. self.image\_prefix=image\_prefix 
40. 
41. **def read(self):** 
42. capture = cv2.VideoCapture(self.videoPath) 
43. win\_name = "test" 
44. cv.NamedWindow(win\_name, cv.CV\_WINDOW\_AUTOSIZE)
45. success,frame = capture.read()
46. 
47. init\_row=0 
48. **while success:** 
49. 
50. self.frame\_count+=1 
51. success,frame = capture.read()
52. \#cv2.imshow(win\_name,frame) \#显示照片浪费时间 
53. time\_duration=self.frameSpan\[init\_row\]\[1\]-self.frameSpan\[init\_row\]\[0\]+1 
54. **if self.frame\_count>=self.frameSpan\[init\_row\]\[0\] and \\** 
55. self.frame\_count<=self.frameSpan\[init\_row\]\[1\] : 
56. 
57. 
58. self.image\_save(frame,init\_row,time\_duration) 
59. 
60. 
61. **if self.frame\_count==self.frameSpan\[init\_row\]\[1\]:** 
62. init\_row+=1 
63. **if init\_row==len(self.frameSpan):** 
64. cv.DestroyWindow(win\_name)
65. **return 0** 
66. 
67. c = cv.WaitKey(10) 
68. **if c == 27:** 
69. **break** 
70. 
71. cv.DestroyWindow(win\_name)
72. **def image\_save(self,frame,init\_row,time\_duration):** 
73. save\_path=self.make\_folder(init\_row,time\_duration) 
74. 
75. \#frame=Image.fromarray(frame) 
76. \#frame.resize((300,300)) 
77. \#frame.save('%s//%s.jpg' % (save\_path,self.frame\_count)) 
78. frame=cv2.resize(frame,(300,300))\# 
79. cv2.imwrite( '%s//%s.jpg' % (save\_path,self.frame\_count), frame) 
80. 
81. 
82. **def make\_folder(self,init\_row,time\_duration):** 
83. temp= self.image\_prefix+'\_'+str(init\_row)+'\_'+str(time\_duration)\#事件名称\_事件顺序\_ 
84. new\_path = os.path.join(self.save\_path,temp) 
85. **if not os.path.isdir(new\_path):** 
86. os.makedirs(new\_path)
87. **return new\_path** 
88. 
89. 
90. 
91. **def read\_excel\_single(excel\_path,event\_instance):** 
92. 
93. data=\[\]
94. data=xlrd.open\_workbook(excel\_path)
95. \#read the first sheets 
96. table = data.sheets()\[0\] 
97. \#read the num fo cols 
98. nrows = table.nrows
99. rowmsg=\[\]\#用来存储帧的范围，及事件的类别，描述。 
100. 
101. **for i in xrange(nrows):** 
102. start\_frame,end\_frame=table.row\_values(i)\[1\].split(':') 
103. rowmsg.append(\[int(start\_frame),int(end\_frame),table.row\_values(i)\[3\]\]) 
104. 
105. rowmsg.sort(**lambda x,y:cmp(x\[0\],y\[0\]))\#按照数字大小排序~** 
106. 
107. event\_id=\[\]
108. event\_id=event\_instance
109. \#print event\_id 
110. 
111. 
112. event=\[\]
113. **for item in xrange(len(rowmsg)):** 
114. **if rowmsg\[item\]\[2\]==event\_id:\#** 
115. event.append(rowmsg\[item\])
116. **if event==\[\]:** 
117. **print ("there is no %s occurs" % event\_instance)** 
118. **return 0** 
119. \#去除重复 
120. 
121. cur\_pos=1 
122. new\_rowmsg=\[\]
123. pre\_s=event\[0\]\[0\] 
124. pre\_e=event\[0\]\[1\] 
125. cur\_s=event\[cur\_pos\]\[0\] 
126. cur\_e=event\[cur\_pos\]\[1\] 
127. 
128. **while cur\_pos<len(event)-1:** 
129. **while not (cur\_s>pre\_e):** 
130. **if cur\_s>pre\_s:** 
131. cur\_s=pre\_s
132. **if cur\_e<pre\_e:** 
133. cur\_e=pre\_e
134. still\_pos=\[cur\_s,cur\_e,event\_id\]
135. cur\_pos+=1 
136. **if cur\_pos==len(event)-1:** 
137. **break** 
138. 
139. pre\_s=cur\_s
140. pre\_e=cur\_e
141. cur\_s=event\[cur\_pos\]\[0\] 
142. cur\_e=event\[cur\_pos\]\[1\] 
143. new\_rowmsg.append(still\_pos)
144. **if cur\_pos==len(event)-1:** 
145. **break** 
146. 
147. cur\_pos+=1 
148. pre\_s=cur\_s
149. pre\_e=cur\_e
150. cur\_s=event\[cur\_pos\]\[0\] 
151. cur\_e=event\[cur\_pos\]\[1\] 
152. 
153. 
154. **return new\_rowmsg** 
155. 
156. 
157. 
158. excel\_source='E:/08ann/dev\_20071101/LGW\_20071101\_E1\_CAM1.mpeg.xlsx' 
159. event\_instance='CellToEar' 
160. 
161. row\_msg=read\_excel\_single(excel\_source,event\_instance)
162. new\_row=\[ item\[0:2\] **for item in row\_msg \]** 
163. 
164. filename = "G:/TrecvidData/08/DEV/LGW\_20071101\_E1\_CAM1.mpg" 
165. 
166. 
167. 
168. videoread=videoFramesExtractor(filename,new\_row,'CellToEar') 
169. videoread.read()

有几点发现：

cv2.的函数输出的数据类型基本都是numpy，而cv.不行。所以尽量在[Python][]中使用cv2.的函数。

通过numpy转换后的图片会在颜色上有些失真，不过影响不大，如果是灰度图的话则完全没有影响。

[20180409103820850]: /images/20220527/21357861af8344c4ba995cdf03584928.png
[SouthEast]: /images/20220527/bcb72a0810aa4274a88d80e388fb9ca8.png
[view plain]: http://blog.csdn.net/vola9527/article/details/38307053#
[Python]: http://lib.csdn.net/base/11