Video motion tracking and speed estimation

On a sunny day (Sun, 15 Mar 2009 09:51:12 -0500) it happened "Anthony Fremont" wrote in :

I am sort of interested, but do you have any other site then a newsgroup that is not available?

"... unwanted words": IEEE WIPO patent portal

...Jim Thompson

I'll upload a video and some individual JPEG images to Photobucket or YouTube in a little while. I'll post the link here for you.

ed

mera.

I just noticed this course on machine vision last night. The complete course lectures are included.

-- Joe

Notice the available translation. We certainly want to make sure the Iranians can make accurately-guided missiles when they are directed at Israel :-(

...Jim Thompson

Here is a link to a folder on photobucket. Sorry about the adds, but it's free.

There are 10 still images and a couple of videos. One of the videos is a much earlier trial.

On a sunny day (Sun, 15 Mar 2009 10:59:55 -0500) it happened "Anthony Fremont" wrote in :

Very nice. The speed measurement depends on the distance I suppose? How do you get the distance? Could perhaps be done with 2 cameras, the parallax would tell you something about it.

It's really just an estimate. It's reasonably accurate, but still just an estimate. The true way to tell how fast they are going is to watch how long it takes to cover a known distance (like you didn't already know that). It's true that I'm giving the further cars a bit of a break on the speed measurement, but I'm far enough back and the street is narrow enough that the distance variation is very limited in range so I don't factor it in.

I basically measure how fast the vertical yellow segments progress thru the frame. Each pixel moved per frame is approximately 1.8MPH. Frames arrive

On a sunny day (Sun, 15 Mar 2009 13:31:41 -0500) it happened "Anthony Fremont" wrote in :

Neat. How do you get the motion vector? I have some motion detection here,

I use subtraction like you do, but I have a couple of different ways of doing it. Each has its strengths and weaknesses. I find "blobs" of change and draw a box around the largest "blob". The rate of change of the X coordinate of the vertical line segments is the motion vector.

That's the part I knew was missing. What is the capture card? Came with the camera, or part of your video card?

It's allot like this card, only by a real noname company.

"Linux Support: Yes"

I like it.

ffmpeg even has a win32 executable. I use it with PSPVideo9 for making a flic for my PSP from one of my DVDs.

More like "Linux Support: Exclusively". I don't do windohs.

I recompiled it with SSE3 support and it went from being able to create the movie at about 30fps, to 180fps. Eats up the CPU though. ffmpeg is single threaded so it usually consumes about 100% of one core when it's running and the captures each eat about 8% of the other core. BTW this is all at

On a sunny day (Sun, 15 Mar 2009 14:10:56 -0500) it happened "Anthony Fremont" wrote in :

I see that you use an expensive frame grabber card. The mcamip I wrote uses ethernet... But.. some years ago I bought some cheap USB camera (Q-Tec), maybe was about 12 $.

Edge detection algorithms?

Better accuracy means military gets hit more often, instead of civilians. Which is good or bad depending on your point of view.

Everything is about motive and opportunity, though, and I'm more concerned about the ease with which bio-warfare is achievable now with hardly more than a kitchen and a pressure cooker and what you can pick up in your backyard. At least, we can keep the CIA focused on people with the opportunity to make missles with some chance of modest success. With bio, everyone has the opportunity if they have a bachelor's worth of training and a few pots and pans. Since almost everyone has the opportunity, the CIA is left with looking only for motive. A much bigger and more varied set.

I remain glad to see MIT's continued commitment to providing high quality education and to make it as accessible as reasonably possible. The other option is to have an elite in control of who gets to have knowledge -- and I consider that worse than most things, with a few exceptions admitted.

Jon

I didn't pay $149 for mine. I got it from newegg and I paid like $50 for it. I really got a hell of a deal since you can't touch a Geovision equivalent for less than $300-$400. Plus I have four audio inputs, if I ever figure out how to enable them.

I've got a d-link ip camera here somewhere. It likes to crash too often, but it is pretty neat with pan and tilt control. It streams mpeg4.

I've got an ibmcam USB camera upstairs somewhere, there's a driver for it still. The nice thing about those cameras is that they are progressive scan. The analog cameras are interlaced with each field exposed at a different time. This makes for really ugly comb like interlace lines in the picture. I had to fix those too as you can see or the faster vehicles look like a double exposure.

;-) BTDT Isn't it wonderful?

I sent an e-mail to the local police chief with the 10 photos up on photobucket.com. Notice the white truck with the stripes on the side doing

You see my mailbox out there......

I can get YUV from the linux driver too, but I went with RGB instead. YUV might have advantages that I could use.

Not sure what you mean. I don't have anything I'd really call an edge detector in my code, just blob analyzers. I use ffmpeg just to make movies out of jpegs.