Path: news.mathworks.com!newsfeed-00.mathworks.com!newsfeed2.dallas1.level3.net!news.level3.com!postnews.google.com!8g2000hsu.googlegroups.com!not-for-mail
From: ImageAnalyst <imageanalyst@mailinator.com>
Newsgroups: comp.soft-sys.matlab
Subject: Re: human tracking
Date: Tue, 1 Apr 2008 18:53:37 -0700 (PDT)
Organization: http://groups.google.com
Lines: 37
Message-ID: <9488d7ca-866e-49e7-8307-88d3845706d6@8g2000hsu.googlegroups.com>
References: <fstnms$965$1@fred.mathworks.com>
NNTP-Posting-Host: 75.186.67.199
Mime-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable
X-Trace: posting.google.com 1207101217 19577 127.0.0.1 (2 Apr 2008 01:53:37 GMT)
X-Complaints-To: groups-abuse@google.com
NNTP-Posting-Date: Wed, 2 Apr 2008 01:53:37 +0000 (UTC)
Complaints-To: groups-abuse@google.com
Injection-Info: 8g2000hsu.googlegroups.com; posting-host=75.186.67.199; 
User-Agent: G2/1.0
X-HTTP-UserAgent: Mozilla/4.0 (compatible; MSIE 7.0; Windows NT 5.1; .NET CLR 
Xref: news.mathworks.com comp.soft-sys.matlab:460449



On Apr 1, 12:26=A0pm, "Chee Sing Wai" <williams_lib2...@hotmail.com>
wrote:
> Hi, I'm developing an algorithm for pedestrian tracking for
> my final year project using Matlab ver. R2006a.
> Most of the object detection and tracking part were almost
> completed base on background modeling technique and
> background subtraction frame by frame.
> The part that I'm facing problems now is to do the human
> recognition to identify the object tracked is human or not.
> As I'm new to this subject, can anyone please help in
> recommend which technique should be used and how?

---------------------------------------------------------------------------
Could be very difficult.  Could be a Ph.D. project instead of just a
simple final year project.  So you have objects moving in your video
and you need to detect whether they are human versus car, truck, dog,
bird, etc.  You better be able to make a lot of assumptions or it's
going to be very difficult.  For example, let's say you have a small
blob moving across the top of your frame.  How do you know it's a bird
(or a plane) and not a human that is far away?  You can't look for
skin color because the person could be walking away and not showing
any skin.  Maybe you can assume that anything moving in a certain zone
is human (say along a sidewalk) and anything moving in another zone
(say, the street) is a vehicle.  You might say that vehicles move
faster than humans but of course we know that this is not always the
case.  Maybe you could do some sort of shape recognition but this is
going to be very hard.  Maybe you could pick out people based on the
size of the blob, but this assumes everything is about the same
distance from the camera.  For a simple example, if you were looking
at a crosswalk with a limited field of view that you had to monitor,
then you could assume that if its size is in a certain range, and its
speed is in a certain range, then it might be human.  Maybe you could
post a few frames (on a web site somewhere), list your known
parameters, restrictions, and assumptions, and then ask the smart
folks over in sci.image.processing for tips.
Regards,
ImageAnalyst