Identify trees in photo

edit

I am trying to use a camera to help localize my robot in my garden. I want to be able to identify trees as they are the most prominent features.

I am having great difficulty extracting the trees from the background. they seem to have the same colur as the surroundings so I cannot use that so i have been trying to identify the edges.

I have looked at each of RGB and HSV and none seperatly seem to show the edges better than grayscale.

I have used a bilateral filter as it is supposed to blur but maintain the edges and I have used a sobel filter to highlight the edges as it seems to make a better job on this than Cany.

Here is a link to the original test photo. link text

I only take the middle third of the picture to reduce the size and because that part has the treetrunks.

Here is my code:

import cv2
import numpy as np

thresh   = 0
sobel_8u = 0
gray     = 0
img      = 0
kernel   = 0

def changeThreshhold(x):

   global thresh,sobel_8u,gray,img
   ret,thresh = cv2.threshold(sobel_8u,x,255,cv2.THRESH_BINARY )

   img = np.vstack((gray,sobel_8u,thresh))
 
treeName  = 'C:\Users\Alan\Desktop\Trees\tree4.jpg'
imgC      = cv2.imread(treeName,cv2.CV_LOAD_IMAGE_COLOR)
h,w,c     = np.shape(imgC)
partImgC = imgC[h/2:2*h/3,:,:]
 
gry  = cv2.cvtColor(partImgC,cv2.COLOR_BGR2GRAY)
gray = cv2.equalizeHist(gry)
 
blur         = cv2.bilateralFilter(gray,21,75,75)
sobelx64f    = cv2.Sobel(blur,cv2.CV_64F,1,0,ksize=3)
abs_sobel64f = np.absolute(sobelx64f)
sobel_8u     = np.uint8(abs_sobel64f)
 
changeThreshhold(65)
 
cv2.namedWindow('image',cv2.cv.CV_WINDOW_NORMAL)
cv2.createTrackbar('threshhold','image',65,255,changeThreshhold)
 
while(1):
    cv2.imshow('image',img)
    k = cv2.waitKey(1) & 0xFF
    if k == 27:
        break    
 
cv2.destroyAllWindows()
<\code><\pre>
 
The results are here:link text
 
They do not look so bad but when I try to use HoughLines to extract the lines and hence the trees I get a huge amount of rubbish.
 
Can anybody show me a better way.
 
Thanks Alan