#SENSORs# ..help with sensor problems! Thank you to all that can assist

I was watching a Scientific American Frontiers on PBS one night. They were doing a segment on robots. One of the highlights was robotic soccer. One of the teams used robots that were controled by a larger computer that kept track of them with video.

The CMU camera tracks objects. Perhaps you could use a similar system.

chris > Hi guys,

Hi Chris,

That has been considered already but knocked back due to the inability to control lighting levels. Due to the large amount of intelligent lighting and projections in the space we came to the conclusion that there could possibly be issues with the tracking. Also, this is going to be an interactive installation which will have a lifespan of around 5years and requiring little to no maintenance. This is why we were hoping for a sensor matrix of proximity sensors or something. A unit that uses infrared/sonic/motion/laser or whatever to provide data to our server. A visual recognition system would be very costly and require continued support for the client. Thank you for the suggestion, much appreciated.

Regards, Nick.

You might consider orienting IR beams floor-to-ceiling. That'd give you a trigger when a body was in a precise spot, as opposed to passing through a plain.

Hi,

That has also been considered! Problem is, I would need hundreds or even thousands to get the necessary resolution in the interactive pathways. The space is huge; 90m in length and 9m wide. That would be far to costly and moreover, difficult to hide so many sensors. Thanks for the idea though. Cheers

Regards, Nick.

"E. Lee Dickinson" wrote in message news:cgiarb$44p$ snipped-for-privacy@solaris.cc.vt.edu...

This seems like a job for video. Software segmenting a video signal and then triggering on motion is fairly common, and the theory behind it is not complex. You can extend the system to indentify "objects" by position, size, and color as they pass through, though I doubt you will be able to identify individual people unless you get real fancy with face recognition. This science is still in its infancy.

Mount several cameras, all pointing down at the same angle, so there is some overlap between them. Depending on the view angle of each camera you might be able to get by with just 10 cameras. Each would have a nine meter field of vision. You can change the field of vision by changing the focal length of the lens.

Consider the lighting of this hall or tunnel. If underlit or unevenly lit you may need to use infrared, and then you won't be able to use color as a differentiator. Avoid going this route. Color makes gross vision analysis much easier.

For this type of application, video is very cheap. It's all in software.

-- Gord> I know this is a robotics ng but I was hoping you could still provide some

What resolution do you need? If you knew someone was in a 1.5 x 1.5 meter square would that be sufficient? Do you have to know that there are two people (as opposed to one) in a grid area? What are the budget constraints - $1000, $10,000, more, less??

Mark

Hi Mark,

We have no budget as yet so I can't say! $8K was what we estimated toward the sensor equipment but this could dbl or half at this point as we have no basis to go on and no options to compare against. The area (tunnel) is 90m in length, 9m wide, and 10m high. I have not considered a sensor segmentation of the space itself as I was hoping to have a sensor (like proximity) that could cover an area of say 10mx10m and then some how using a combination of software and triangulation know how far the target is in relation to itself (the sensor) and where in the space (global) it is. This would allow me to cover the entire space using the most minimal sensor units possible. However, I could be in a dream world as I don't know enough about what is out there and available. If I had to give the optimal resolution of the area or sensor hotspots for interactivity purposes, then I would say somewhere in the vicinity of 3mx1.8m but this would require 150 sensors which is out of the question! The hotspots could be doubled decreasing the required sensors also by half but this is still far to many. Any other suggestions would be great. Cheers!

Regards, Nick.

Hi Gordon,

Can you elaborate more on vision via camera tracking? The space at the peak of interactivity i.e. all devices running at a signal time would simulate an environment like a rave or dance party (not sure what you call them from where your from). So the cameras would need to identify moving objects in a dark environment with changing light. There is no need for camera recognition, just identify an object in the space and send the server a coord so it can then activate a series of devices in the targets location.

This is a public space so there is a lighting level to which the space cannot go below for safety reasons so this may save us if we use camera tracking technology. The main concern is your statement "This science is still in its infancy" - this could require a large amount of maintenance which would be a negative option for us. If you have any more information, I would greatly appreciate it. Cheers!

Regards, Nick.

You really need to 'look down' on things to get the least cluttered images - fortunately you have some headroom. Video is probably the only realistic solution, but you need to give it as much help as possible. If you cover the floor with a distinctive pattern carpet, then you can inspect the video images for 'holes' in the pattern, which hopefully are the people you need to follow. Colour might work, but runs the risk of being very confused by shadows and different coloured clothes.

Your budget is hopelessly inadequate !

Dave

constraints -

Posted Via Nuthinbutnews.Com Premium Usenet Newsgroup Services

---------------------------------------------------------- ** SPEED ** RETENTION ** COMPLETION ** ANONYMITY **

----------------------------------------------------------

Inadequate for what? Someone may yet come up with a clever $20 solution.

You are right of course - I shouldn't be so negative, and in fact I have two sub $20 plans already !

2) Train pigeons to peck at the images on a tv screen

:>) Dave

Posted Via Nuthinbutnews.Com Premium Usenet Newsgroup Services

---------------------------------------------------------- ** SPEED ** RETENTION ** COMPLETION ** ANONYMITY **

----------------------------------------------------------

It is not to difficult. Just use b/w camera's with IR filters in front and IR lights (LED's flashers). For 90 meter tunnel you need something like 10 camera's on the sealing. Give each camera it's own computer and put then on a network. They all report to one server. Tracking individuals in a group is difficult, identifing if an object is at a specific position not.

Good luck,

Peter

What does the floor look like? Is there a regular tiling or a distinctive color or pattern? If so, then you could use overhead cameras to watch for human sized blobs of pixels that don't match the pattern.

Since you have a very high ceiling (10 meters ~= 33 feet) each camera should be able to cover a lot of area. You might be able to watch the entire area with 3 or 4 cameras.

Especially if you consider the floor pattern a constant. You might want to assume that any change that remains constant for a certain amount of time is part of the "fixed" background. That way you don't have to reprogram if somebody moves a trash can.

Agreed. All the other solutions I've thought of either won't work or are much less workable.

-- D. Jay Newman

I'm not Gordon, but I'll answer anyway.

Assume the cameras don't move and are looking down onto the floor.

A computer has an image of the floor with *nobody* there.

While running, do color correction and then compare the image with the stored background.

Anyplace where the image is sufficiently different from the stored background is a person.

Anytime a section of the image remains constant, assume that something has been moved and add this to the background.

I might suggest that you should have started worrying about the sensors quite some time ago, but I'm a nice person so I won't.

The only other suggestion I can make is a pressure-sensitive carpet, but that would cost a small fortune. That is, a small fortune for Bill Gates. :)

-- D. Jay Newman

I am unclear what it is you are looking for. What is an "object"? I assume you don't have control over the shape or coloration of the objects. Are they people? What do you mean by "simulate an environment"? Isn't this a real-world situation you're developing for?

Do you need to *identify* objects to track them in individual trajectories, or is it sufficient just to determine if an object occupies some space, and activate something in response? The latter is doable. The former is not practical given the number of "objects" comparable to a rave party. I don't think anyone has come up with this level of image tracking.

I understand that you may be unable to freely discuss what you're planning, but it's not really possible to provide specific answers without specific detail.

-- Gordon Author: Robot Builder's Bonanza Budget Robotics:

What are the rules? - Are you allowed to tag the targets - have them wear badges with RFIDs for example?

Kevin

The nature of the interaction is unclear. Surely the sensors would be in whatever the target is to interact with? Why is there a need to track the people?

Hi Nick, here's my idea, but you owe me royalties when you use it - [just kidding, I suppose posting it puts it in the public domain - meaning you prolly cannot patent the idea later on] ....

How about multiple CCD cams overhead, pointing straight down, ID'ing certain people by their size, clothing color, direction/speed of movement, etc, and then following them while they move from the viewing area of one camera to the next. Each cam will view maybe a 20' by 20' limited space, meaning at most you will have 20 to 30 moving objects in each frame to keep track of. Miscellaneous fixed objects [litter cans, chairs, etc] will not move, and will be subtracted out between frames of a given cam.

This is better than the scheme they used in the Minority Report movie, which was to perform eye/retinal scans via laser every 10' or so to track Tom Cruise as he moved along - at least as far as personal invasion goes. BTW, if you can get the people to wear license plates, then the UK already has a similar scheme in operation on their autoroutes. Also, you might check on the college soccer competitions, as they use something similar to the overhead cam thingie.

Aside from this, you might also check through the dozens of projects undertaken at the MIT Media Lab, as this sounds like something they'd have tried already.

- dan michaels