1
Meanshift and Camshift {#tutorial_py_meanshift}
9
- We will learn about Meanshift and Camshift algorithms to find and track objects in videos.
14
The intuition behind the meanshift is simple. Consider you have a set of points. (It can be a pixel
15
distribution like histogram backprojection). You are given a small window ( may be a circle) and you
16
have to move that window to the area of maximum pixel density (or maximum number of points). It is
17
illustrated in the simple image given below:
19
![image](images/meanshift_basics.jpg)
21
The initial window is shown in blue circle with the name "C1". Its original center is marked in blue
22
rectangle, named "C1_o". But if you find the centroid of the points inside that window, you will
23
get the point "C1_r" (marked in small blue circle) which is the real centroid of window. Surely
24
they don't match. So move your window such that circle of the new window matches with previous
25
centroid. Again find the new centroid. Most probably, it won't match. So move it again, and continue
26
the iterations such that center of window and its centroid falls on the same location (or with a
27
small desired error). So finally what you obtain is a window with maximum pixel distribution. It is
28
marked with green circle, named "C2". As you can see in image, it has maximum number of points. The
29
whole process is demonstrated on a static image below:
31
![image](images/meanshift_face.gif)
33
So we normally pass the histogram backprojected image and initial target location. When the object
34
moves, obviously the movement is reflected in histogram backprojected image. As a result, meanshift
35
algorithm moves our window to the new location with maximum density.
37
### Meanshift in OpenCV
39
To use meanshift in OpenCV, first we need to setup the target, find its histogram so that we can
40
backproject the target on each frame for calculation of meanshift. We also need to provide initial
41
location of window. For histogram, only Hue is considered here. Also, to avoid false values due to
42
low light, low light values are discarded using **cv2.inRange()** function.
47
cap = cv2.VideoCapture('slow.flv')
49
# take first frame of the video
50
ret,frame = cap.read()
52
# setup initial location of window
53
r,h,c,w = 250,90,400,125 # simply hardcoded the values
54
track_window = (c,r,w,h)
56
# set up the ROI for tracking
57
roi = frame[r:r+h, c:c+w]
58
hsv_roi = cv2.cvtColor(roi, cv2.COLOR_BGR2HSV)
59
mask = cv2.inRange(hsv_roi, np.array((0., 60.,32.)), np.array((180.,255.,255.)))
60
roi_hist = cv2.calcHist([hsv_roi],[0],mask,[180],[0,180])
61
cv2.normalize(roi_hist,roi_hist,0,255,cv2.NORM_MINMAX)
63
# Setup the termination criteria, either 10 iteration or move by atleast 1 pt
64
term_crit = ( cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 10, 1 )
67
ret ,frame = cap.read()
70
hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
71
dst = cv2.calcBackProject([hsv],[0],roi_hist,[0,180],1)
73
# apply meanshift to get the new location
74
ret, track_window = cv2.meanShift(dst, track_window, term_crit)
77
x,y,w,h = track_window
78
img2 = cv2.rectangle(frame, (x,y), (x+w,y+h), 255,2)
79
cv2.imshow('img2',img2)
81
k = cv2.waitKey(60) & 0xff
85
cv2.imwrite(chr(k)+".jpg",img2)
90
cv2.destroyAllWindows()
93
Three frames in a video I used is given below:
95
![image](images/meanshift_result.jpg)
100
Did you closely watch the last result? There is a problem. Our window always has the same size when
101
car is farther away and it is very close to camera. That is not good. We need to adapt the window
102
size with size and rotation of the target. Once again, the solution came from "OpenCV Labs" and it
103
is called CAMshift (Continuously Adaptive Meanshift) published by Gary Bradsky in his paper
104
"Computer Vision Face Tracking for Use in a Perceptual User Interface" in 1988.
106
It applies meanshift first. Once meanshift converges, it updates the size of the window as,
107
\f$s = 2 \times \sqrt{\frac{M_{00}}{256}}\f$. It also calculates the orientation of best fitting ellipse
108
to it. Again it applies the meanshift with new scaled search window and previous window location.
109
The process is continued until required accuracy is met.
111
![image](images/camshift_face.gif)
113
### Camshift in OpenCV
115
It is almost same as meanshift, but it returns a rotated rectangle (that is our result) and box
116
parameters (used to be passed as search window in next iteration). See the code below:
121
cap = cv2.VideoCapture('slow.flv')
123
# take first frame of the video
124
ret,frame = cap.read()
126
# setup initial location of window
127
r,h,c,w = 250,90,400,125 # simply hardcoded the values
128
track_window = (c,r,w,h)
130
# set up the ROI for tracking
131
roi = frame[r:r+h, c:c+w]
132
hsv_roi = cv2.cvtColor(roi, cv2.COLOR_BGR2HSV)
133
mask = cv2.inRange(hsv_roi, np.array((0., 60.,32.)), np.array((180.,255.,255.)))
134
roi_hist = cv2.calcHist([hsv_roi],[0],mask,[180],[0,180])
135
cv2.normalize(roi_hist,roi_hist,0,255,cv2.NORM_MINMAX)
137
# Setup the termination criteria, either 10 iteration or move by atleast 1 pt
138
term_crit = ( cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 10, 1 )
141
ret ,frame = cap.read()
144
hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
145
dst = cv2.calcBackProject([hsv],[0],roi_hist,[0,180],1)
147
# apply meanshift to get the new location
148
ret, track_window = cv2.CamShift(dst, track_window, term_crit)
151
pts = cv2.boxPoints(ret)
153
img2 = cv2.polylines(frame,[pts],True, 255,2)
154
cv2.imshow('img2',img2)
156
k = cv2.waitKey(60) & 0xff
160
cv2.imwrite(chr(k)+".jpg",img2)
165
cv2.destroyAllWindows()
168
Three frames of the result is shown below:
170
![image](images/camshift_result.jpg)
175
-# French Wikipedia page on [Camshift](http://fr.wikipedia.org/wiki/Camshift). (The two animations
177
2. Bradski, G.R., "Real time face and object tracking as a component of a perceptual user
178
interface," Applications of Computer Vision, 1998. WACV '98. Proceedings., Fourth IEEE Workshop
179
on , vol., no., pp.214,219, 19-21 Oct 1998
184
-# OpenCV comes with a Python sample on interactive demo of camshift. Use it, hack it, understand