Detect a human figure in a live video

I am new to matalb,I want to detect a moving human figure in live video feed, this human figure can be in different poses. Can anyone give me some advice on how to detect the human figure after detecting its(human figure) motion.

Thanks,

[Milindu]

http://www.di.ens.fr/~rodrigue/human_detection.html

Seems too complex

[Milindu]

Gait recognition is based on most of the human detection system, still it seems that it has many flows, I'm still searching. If anyone has more info on gait recognition.

[Image Analyst]

There is information on gait analysis in the Iris USC bibliography I gave you below. See section "16.7.4.5 Walking, Gait Recognition, Gait Analysis". But you're going to have to do a LOT of work on that complicated subject. It's nothing where we can give you a hundred lines of code and you're all set.

Answer 1

[Stephen]

if the camera is static, take a blank image and subtract it from the live feed to get a rough idea of what pixel values have changed. maybe division would handle different daylight conditions better. anyway, once you have something moving or different about the picture, you can do a medial transform of the shape to see how 'stick figure'-like it is. good luck

Answer 2

[Image Analyst]

Search the File Exchange for "tracking" or see this Mathworks webinar: http://www.mathworks.com/matlabcentral/fileexchange/35646-march-2012-demo-files-for-computer-vision-with-matlab

[Image Analyst]

Then you need to look up algorithms listed in section 16 or 18 here: http://iris.usc.edu/Vision-Notes/bibliography/contents.html

[Image Analyst]

There is a conference next month with sessions on gait analysis if you're interested: http://hhp.ufl.edu/apk/ces/ASB2012/2012%20ASB%20conference%20program%20schedule.htm

Answer 3

[Walter Roberson]

Gait recognition is not sufficient for your purpose. Consider people on crutches, people in wheelchairs, people on unicycles, bicycles, or tricycles. Some of the robots being built in Japan apparently have quite good bipedal motion and look quite realistic, but those are not human.

Recognizing leg length accurately in order to calculate proportions in order to distinguish humans from other primates (e.g., gorillas) is difficult when one considers skirts, dresses, and long winter coats. The proportions in humans vary considerably -- consider for example dwarfism. Even without disorders such as that, if you take a male human and a female human of the same height, the female will statistically have longer legs. And I know that my proportions are at least two standard deviations from the mean.

For the purpose of this project, are you restricting "human" to Homo sapiens sapiens, or are you including Homo neanderthalensis, Homo habilis, and so on?

Answer 4

[Milindu]

Tnx

There are many information on tracking in there but i can not find any information on how to detect a human figure, Actually human figure can have different posses, but if i take a walking human the legs are always changing in a certain way can i take this action into consideration and deduce because of the motion part it is a human.

[Image Analyst]

See my comment I added to my answer. By the way, this post of yours is not an "Answer" to your question - it should have been added as a comment on my Answer, so that's why I answered this up there instead of here.

[Milindu]

my bad

Answer 5

[Geoff]

Employ a 2 year-old child to sit in front of the screen and "point at the person".

This solution is 100% accurate with absolutely no mathematics required. It works with occlusion, non-static background, non-static camera, image noise, poor lighting, and multiple targets.

That's how frustrating Computer Vision is!!!

[Milindu]

I wish i can submit your answer...it really is frustrating

[Walter Roberson]

Does the 2 year old identify Barney as human? How about Sesame Street's Bert? Barney has a human inside the costume, so should be identified as human, but Bert is always a puppet and so should not.

Okay, now how about R2D2? Some versions of R2D2 have humans inside and so should be identified as human, but some of them are electromechanical. Can a 2 year old tell the difference? Likewise for the new Daleks, the authorized commercial versions of which are electromechanical, and some of the background Daleks in crowd scenes were electromechanical, but the close-ups in the Dr. Who series are only of Daleks with humans inside.

[Geoff]

Are you suggesting that a computer-based algorithm should also be able to deduce that the purple dinosaur or protocol droid actually has a human inside?

[Walter Roberson]

I am suggesting that the 2 year-old child is not "100% accurate", that there are entire categories of situations in which they are likely to be wrong. I would not suggest that a computer based algorithm would be any better, but we need to understand the limitations of our solutions.

[Geoff]

I imagine that the telling of jokes at your dinner table must be quite a spectacle! =)

[Milindu]

As longs as Barney and Bert wouldn't come out of the Sesame Street and R2D2 is still in creation processes i hope i would be safe.......

What ever the process there will be lot of limitations, even with them i don't think computer based algorithms would be able to archive 10% accuracy the 2-year archive with monitoring..

[Walter Roberson]

So you are designing your security system to fail at comic book and science fiction conventions?

[Geoff]

If this is for a 24/7 system, be aware that 2 year-old humans don't have reliable up-time.

[Walter Roberson]

And if you have a 2-year old human riding a cow, such as are shown in this security snapshot provided by "nomad nomad" ? http://www.casimages.com/img.php?i=120531055451175771.png