有人对我如何在python中进行图像比较以检测图像中的更改有什么建议吗?我目前正在开发一个应用程序,它将用我的摄像头监控我的区域,我想知道如何比较每帧拍摄的图像,看看是否检测到任何运动。从长远来看,我想设置一个敏感度滑块,所以如果你能引导我的方向,我相信我可以解决其余的问题。
正如我在这里看到的一些帖子询问如何将网络摄像头与wxPython集成,这里有一个小演示。请注意,我昨天晚上才开始,所以如果你想找提示顶部代码,你可能需要自己修改它(目前;):
要求:PIL&;VideoCapture
#videocapturepanel.py
#Todo:
# - Fix background colour after video is stopped
# - Create image comparison method
# - Add capture function
# - Save stream to video file?
import threading, wx
from PIL import Image
from VideoCapture import Device
cam = Device(0)
buffer, width, height = cam.getBuffer()
cam.setResolution(width, height)
DEFAULT_DEVICE_INDEX = 0
DEFAULT_DEVICE_WIDTH = width
DEFAULT_DEVICE_HEIGHT = height
DEFAULT_BACKGROUND_COLOUR = wx.Colour(0, 0, 0)
class VideoCaptureThread(threading.Thread):
def __init__(self, control, width=DEFAULT_DEVICE_WIDTH, height=DEFAULT_DEVICE_HEIGHT, backColour=DEFAULT_BACKGROUND_COLOUR):
self.backColour = backColour
self.width = width
self.height = height
self.control = control
self.isRunning = True
self.buffer = wx.NullBitmap
threading.Thread.__init__(self)
def getResolution(self):
return (self.width, self.height)
def setResolution(self, width, height):
self.width = width
self.height = height
cam.setResolution(width, height)
def getBackgroundColour(self):
return self.backColour
def setBackgroundColour(self, colour):
self.backColour = colour
def getBuffer(self):
return self.buffer
def stop(self):
self.isRunning = False
def run(self):
while self.isRunning:
buffer, width, height = cam.getBuffer()
im = Image.fromstring('RGB', (width, height), buffer, 'raw', 'BGR', 0, -1)
buff = im.tostring()
self.buffer = wx.BitmapFromBuffer(width, height, buff)
x, y = (0, 0)
try:
width, height = self.control.GetSize()
if width > self.width:
x = (width - self.width) / 2
if height > self.height:
y = (height - self.height) / 2
dc = wx.BufferedDC(wx.ClientDC(self.control), wx.NullBitmap, wx.BUFFER_VIRTUAL_AREA)
dc.SetBackground(wx.Brush(self.backColour))
dc.Clear()
dc.DrawBitmap(self.buffer, x, y)
except TypeError:
pass
except wx.PyDeadObjectError:
pass
self.isRunning = False
class VideoCapturePanel(wx.Panel):
def __init__(self, parent, id=-1, pos=wx.DefaultPosition, size=wx.DefaultSize, initVideo=False, style=wx.SUNKEN_BORDER):
wx.Panel.__init__(self, parent, id, pos, size, style)
if initVideo:
self.StartVideo()
self.Bind(wx.EVT_CLOSE, self.OnClose)
def OnClose(self, event):
try:
self.Device.stop()
except:
pass
def StopVideo(self):
self.Device.stop()
self.SetBackgroundColour(self.Device.backColour)
dc = wx.BufferedDC(wx.ClientDC(self), wx.NullBitmap)
dc.SetBackground(wx.Brush(self.Device.backColour))
dc.Clear()
def StartVideo(self):
self.Device = VideoCaptureThread(self)
self.Device.start()
def GetBackgroundColour(self):
return self.Device.getBackgroundColour()
def SetBackgroundColour(self, colour):
self.Device.setBackgroundColour(colour)
class Frame(wx.Frame):
def __init__(self, parent, id=-1, title="A Frame", path="", pos=wx.DefaultPosition, size=wx.DefaultSize, style=wx.DEFAULT_FRAME_STYLE):
wx.Frame.__init__(self, parent, id, title, pos, size, style)
self.VidPanel = VideoCapturePanel(self, -1, initVideo=False)
self.StartButton = wx.ToggleButton(self, -1, "Turn On")
self.ColourButton = wx.Button(self, -1, "Change Background")
szr = wx.BoxSizer(wx.VERTICAL)
bszr = wx.BoxSizer(wx.HORIZONTAL)
bszr.Add(self.StartButton, 0, wx.ALIGN_CENTER_HORIZONTAL | wx.LEFT, 5)
bszr.Add(self.ColourButton, 0, wx.ALIGN_CENTER_HORIZONTAL)
szr.Add(self.VidPanel, 1, wx.EXPAND)
szr.Add(bszr, 0, wx.ALIGN_CENTER_HORIZONTAL)
self.SetSizer(szr)
self.StartButton.Bind(wx.EVT_TOGGLEBUTTON, self.OnToggled)
self.ColourButton.Bind(wx.EVT_BUTTON, self.OnColour)
def OnColour(self, event):
dlg = wx.ColourDialog(self)
dlg.GetColourData().SetChooseFull(True)
if dlg.ShowModal() == wx.ID_OK:
data = dlg.GetColourData()
self.VidPanel.SetBackgroundColour(data.GetColour())
dlg.Destroy()
def OnToggled(self, event):
if event.IsChecked():
self.VidPanel.StartVideo()
else:
self.VidPanel.StopVideo()
#self.VidPanel.SetBackgroundColour(data.GetColour())
if __name__ == "__main__":
# Run GUI
app = wx.PySimpleApp()
frame = Frame(None, -1, "Test Frame", size=(800, 600))
frame.Show()
app.MainLoop()
del app
*
更新*
使用Paul的示例,我创建了一个类并将其实现到我的代码中:
class Images:
def __init__(self, image1, image2, threshold=98, grayscale=True):
self.image1 = image1
if type(image1) == str:
self.image1 = Image.open(self.image1)
self.image2 = image2
if type(image2) == str:
self.image2 = Image.open(image2)
self.threshold = threshold
def DoComparison(self, image1=None, image2=None):
if not image1: image1 = self.image1
if not image2: image2 = self.image2
diffs = ImageChops.difference(image1, image2)
return self.ImageEntropy(diffs)
def ImageEntropy(self, image):
histogram = image.histogram()
histlength = sum(histogram)
probability = [float(h) / histlength for h in histogram]
return -sum([p * math.log(p, 2) for p in probability if p != 0])
然后将变量self.image=False添加到VideoCaptureAread的__init__()
函数中,并将下面的代码添加到VideoCaptureAread的run()函数的行im=image.fromstring(…)之后:
if self.image:
img = compare.Images2(im, self.image).DoComparison()
print img
self.image = im
当我运行示例时,它看起来工作正常,但我对得到的结果有点困惑:
1.58496250072
5.44792407663
1.58496250072
5.44302784225
1.58496250072
5.59144486002
1.58496250072
5.37568050189
1.58496250072
到目前为止,似乎每一个图像都关闭了相当一点,虽然变化是最小的?理论上,要运行的附加程序应该捕获变量self.image下的先前图像,并与新图像进行比较。在比较之后,使用self.image=im将self.image更新到当前图像,那么为什么每秒钟的图像都会有这样的差异?最多我的眼睛可能在这两张照片里前后移动,我看不出这会对我的结果造成如此大的影响?
*
更新2*
到目前为止,有三个比较比较类和三种不同的运动检测方法。
类图像~我在谷歌搜索时发现的第一次尝试使用一些代码,甚至记不起它是如何工作的。:P页
类Images2~使用此线程中的Paul代码创建,实现其更新的图像熵函数。
类Images3~找到了DetectMotion函数的修改版本here。(返回已更改的百分比,并且似乎考虑了照明)
事实上,我真的不知道他们中的任何一个在做什么,字面上,但我能告诉你的是,到目前为止,类Image3似乎是设置检测的最简单/准确的方法,失败的是它比其他两个类需要更多的时间来处理。
(请注意,为了避免与scipy冲突,进行了一些导入更改,sys.modules[“Image”]与PIL.Image相同)
import math, sys, numpy as np
import PIL.Image, PIL.ImageChops
sys.modules["Image"] = PIL.Image
sys.modules["ImageChops"] = PIL.ImageChops
from scipy.misc import imread
from scipy.linalg import norm
from scipy import sum, average
DEFAULT_DEVICE_WIDTH = 640
DEFAULT_DEVICE_HEIGHT = 480
class Images:
def __init__(self, image1, image2, threshold=98, grayscale=True):
if type(image1) == str:
self.image1 = sys.modules["Image"].open(image1)
self.image2 = sys.modules["Image"].open(image2)
if grayscale:
self.image1 = self.DoGrayscale(imread(image1).astype(float))
self.image2 = self.DoGrayscale(imread(image2).astype(float))
else:
self.image1 = imread(image1).astype(float)
self.image2 = imread(image2).astype(float)
self.threshold = threshold
def DoComparison(self, image1=None, image2=None):
if image1: image1 = self.Normalize(image1)
else: image1 = self.Normalize(self.image1)
if image2: image2 = self.Normalize(image2)
else: image2 = self.Normalize(self.image2)
diff = image1 - image2
m_norm = sum(abs(diff))
z_norm = norm(diff.ravel(), 0)
return (m_norm, z_norm)
def DoGrayscale(self, arr):
if len(arr.shape) == 3:
return average(arr, -1)
else:
return arr
def Normalize(self, arr):
rng = arr.max()-arr.min()
amin = arr.min()
return (arr-amin)*255/rng
class Images2:
def __init__(self, image1, image2, threshold=98, grayscale=True):
self.image1 = image1
if type(image1) == str:
self.image1 = sys.modules["Image"].open(self.image1)
self.image2 = image2
if type(image2) == str:
self.image2 = sys.modules["Image"].open(image2)
self.threshold = threshold
def DoComparison(self, image1=None, image2=None):
if not image1: image1 = self.image1
if not image2: image2 = self.image2
diffs = sys.modules["ImageChops"].difference(image1, image2)
return self.ImageEntropy(diffs)
def ImageEntropy(self, image):
w,h = image.size
a = np.array(image.convert('RGB')).reshape((w*h,3))
h,e = np.histogramdd(a, bins=(16,)*3, range=((0,256),)*3)
prob = h/np.sum(h)
return -np.sum(np.log2(prob[prob>0]))
def OldImageEntropy(self, image):
histogram = image.histogram()
histlength = sum(histogram)
probability = [float(h) / histlength for h in histogram]
return -sum([p * math.log(p, 2) for p in probability if p != 0])
class Images3:
def __init__(self, image1, image2, threshold=8):
self.image1 = image1
if type(image1) == str:
self.image1 = sys.modules["Image"].open(self.image1)
self.image2 = image2
if type(image2) == str:
self.image2 = sys.modules["Image"].open(image2)
self.threshold = threshold
def DoComparison(self, image1=None, image2=None):
if not image1: image1 = self.image1
if not image2: image2 = self.image2
image = image1
monoimage1 = image1.convert("P", palette=sys.modules["Image"].ADAPTIVE, colors=2)
monoimage2 = image2.convert("P", palette=sys.modules["Image"].ADAPTIVE, colors=2)
imgdata1 = monoimage1.getdata()
imgdata2 = monoimage2.getdata()
changed = 0
i = 0
acc = 3
while i < DEFAULT_DEVICE_WIDTH * DEFAULT_DEVICE_HEIGHT:
now = imgdata1[i]
prev = imgdata2[i]
if now != prev:
x = (i % DEFAULT_DEVICE_WIDTH)
y = (i / DEFAULT_DEVICE_HEIGHT)
try:
#if self.view == "normal":
image.putpixel((x,y), (0,0,256))
#else:
# monoimage.putpixel((x,y), (0,0,256))
except:
pass
changed += 1
i += 1
percchange = float(changed) / float(DEFAULT_DEVICE_WIDTH * DEFAULT_DEVICE_HEIGHT)
return percchange
if __name__ == "__main__":
# image1 & image2 MUST be legit paths!
image1 = "C:\\Path\\To\\Your\\First\\Image.jpg"
image2 = "C:\\Path\\To\\Your\\Second\\Image.jpg"
print "Images Result:"
print Images(image1, image2).DoComparison()
print "\nImages2 Result:"
print Images2(image1, image2).DoComparison()
print "\nImages3 Result:"
print Images3(image1, image2).DoComparison()
这可能是一个幼稚的方法,但这是一个简单的开始。我相信你会受到相机噪音的影响,你可能想区分光线的变化和图像构图的变化。但我想到的是:
您可以使用PILImageChops来有效地获取图像之间的差异。然后,可以取diff的entropy来获得单个值阈值。
似乎有效:
我相信这是一个更好的图像熵算法,因为它在颜色空间中进行三维分类,而不是为每个波段创建单独的直方图。
编辑-此功能已于2012年4月6日更改
以下是我的测试图片:
图1
图2
图3
相关问题 更多 >
编程相关推荐