Inspired from the PIL version here
http://gumuz.looze.net/wordpress/index.php/archives/2005/06/06/python-webcam-fun-motion-detection/

First with typical import and Canvas, exit setup

   1  
   2  from appuifw import *
   3  from graphics import Image
   4  import camera, e32
   5  #import miso    # don't dim the light
   6  
   7  app.body = c = Canvas()
   8  
   9  running = 1
  10  def quit():
  11      global running
  12      running = 0
  13  app.exit_key_handler=quit

Then the getdata function that reads pixel data
from image by saving/reading PNG.

   1  
   2  def getdata(im, bpp=24):
   3      import struct, zlib
   4      im.save('D:\\pixels.png', bpp=bpp, compression='no')
   5      f = open('D:\\pixels.png', 'rb')
   6      f.seek(8 +8+13+4)
   7      chunk = []
   8      while 1:
   9          n = struct.unpack('>L', f.read(4))[0]
  10          if n==0: break  # 'IEND' chunk
  11          f.read(4) # 'IDAT'
  12          chunk.append(f.read(n))
  13          f.read(4)   # CRC
  14      f.close()
  15      return zlib.decompress(''.join(chunk))  # '\x00' prefix each line

Lastly, the real code follows.

   1  
   2  last1 = '\x00' * 930    # can be anything
   3  while running:
   4      im = camera.take_photo('RGB', (160,120))
   5      im.rectangle([(10,10),(40,40)], 0xff0000)   # red outline
   6      im.rectangle([(120,10),(150,40)], 0xff0000) # no code for this square
   7      # check hot spot whether active
   8      box = Image.new((30,30), 'L')  # gray scale
   9      box.blit(im, (10,10,40,40))
  10      data = getdata(box, 8)
  11      # check difference for motion
  12      pixdiff = 0
  13      for i in range(len(data)):
  14          if abs(ord(data[i])-ord(last1[i])) > 15:  # pix threshold 15/256
  15              pixdiff += 1
  16              if pixdiff > 90:    # img threshold 90/900
  17                  im.rectangle([(10,10),(40,40)], fill=0xff0000)  # fill
  18                  break           # motion detected
  19      last1 = data
  20      c.blit(im, (0,0), (8,12))   # show camera
  21      #miso.reset_inactivity_time()

When measured, it takes around 1.1 sec for each loop.
Compared this with 0.9 sec without image processing,
out code is quite efficient.