1

I am doing a project on automated power management. My idea is to use a web cam which takes images at regular images and finds out if the room is empty or filled.

Now my idea is that the first image will be of an empty room and there will be a software which will compare the consecutive images with this as base and infer whether the room is empty or not.

I thought pattern recognition would be apt. The software would match the patterns of both images (converted in to binary threshold images probably, its just what I thought) and detect accordingly.

First please tell me if its right (as in is pattern recognition the right step). Secondly, I want to know how to get started on this one

doppelgreener
  • 4,809
  • 10
  • 46
  • 63
Bhagirath N Sai
  • 199
  • 4
  • 16
  • 1
    If there is a large object like a table in the room, and that table is moved, even if no people are in the room the difference will be large and would trigger that the room is occupied. Thermal imaging would be a more accurate method, as mentioned by @HighPerformanceMark – spex Dec 19 '12 at 17:55

2 Answers2

1

If you just want to know if someone is in the room or no (i.e. not trying to count people), you can achieve this by just comparing frame pictures to see if there is a significant change, see this question for many libraries that can help you getting started.

Generally, it is much easier to detect if someone in the room with a PIR sensor (aka motion detection sensor), you can get one from e-bay (less than $10) or local electronic store and a micro-controller such as Arduino and here is a tutorial if you want to go with this method.

Community
  • 1
  • 1
iTech
  • 18,192
  • 4
  • 57
  • 80
0

There are two main problems in practical realization of motion detection algorithms:

  • camera movement (it is impossible, in general, to distinguish camera movement from object movement)

  • compression artefacts (it is hard to tell whether something has moved or it was just a noise)

Since in your situation only the second problem exists I recommend statistical solution presented in this article ("Moving object detection in wavelet compressed video", Toreyin et al., Elsevier):

http://yoksis.bilkent.edu.tr/doi_getpdf/articles/10.1016-j.image.2004.12.002.pdf

I've implemented this solution, and it works quite well even on very noisy videos, if the camera is standing still. You can even implement this paradigm without using wavelet transform and apply this solution just on raw image pixels.

BartoszKP
  • 34,786
  • 15
  • 102
  • 130