# What type of LED lights are needed for detecting retro-reflective markers for motion capture project?

I've a project in which I want to add 10-12 led lights around webcam's lens so that the camera can detect retro-reflective tape in it's viewing area.

How this process works is something like this. I'll power on the red led lights while taking pictures of a human model. The red light of leds will make the captured frames' tapes put on human model light up very brightly(reflection of tapes will be at the same path of led light's emission-because the tape is retro-reflective) so I can separate the tape from other objects in the scene and find its 3d coordinate using c++ programming and image processing software.

I need advice in buying the led lights from http://www.digikey.com. Because I need very bright red led lights and also efficient(low voltage like 1.5 - 2volt and low amperage) what should I look for in the spec? Do I need to change the color of led lights(the tape reflects 50x more than normal white diffused surface)?

What is Millicandela rate?

How this rating help choosing the brightest led light?

The camera's vertical angle of view is 52 degrees and horizontal FOV is 75 degrees. So if I need to create a technically professional looking mocap camera what type of led lights do I need for this project?

If I capture the 2d images in a brightly lit room will the led lights and my process(explained above) work with the camera? What wavelength of led light is important for this project? Should I broaden my spec for voltage and amperage?

If you choose synchronous detection, then you can simplify solution. Then you may want to choose LEDs as narrow angled as you can withing any part of visible spectrum.

Synchronous detection is the way to detect a signal which may be lower than background noise. Imagine taking shot with camera before LEDs are up. Then turn LEDs on and take another picture with exactly same conditions. Subtract arrays of pixel values of 2 frames from each other and look at the remainder. It will contain perfect difference of what was lit by LEDs. In your setup the DIFF picture will contain exactly retroreflectors (cataphote) only. Because the rest of the scenery got no significant increase in brightness.

While you debug the solution it will be obvious that power of LEDs can be reduced to some very low level, but only experiments will tell. I estimate that you will not need any super-powerful diodes. Some ordinary white LEDs with very acurate lens will do.

• This is the correct way to do it, but it is not as easy as what Rocket Surgeon implies. In any video, two sequential frames will always have differences. You will need software to help filter these out and figure out what is "noise" and what is your tape. Also, You need a scheme to synchronize your pulsed LED's and the video capture of your camera. This is basically impossible for consumer cameras, and even some professional cameras. – user3624 Jun 4 '12 at 1:55
• Agree. But who knows what the inventor is trying to accomplish. Be it laser tag game or some military skawking protocol of automatic human killing robot. Knowing the prior art should be always first on the list. I mean the budget can be anywhere in between $10 to$10M, so wiring setup can possibly be flexible. – user924 Jun 4 '12 at 2:02
• Thanks allYes background subtraction is an option. But I'm thresholding the greyscale image. Because there's only white markers in that image. Then find the circles and filter by radius and area. About the synchronization I'm running the cameras at 100Hz with 8 cameras. So i'm basically taking 100 frames per 1000millisecond per camera(I know it's possible! with ps3 eye camera @320x240 image) I know it's off topic but do you think synchronization will be an issue here? Because the fastest part of human body is eyelid. Even that takes 300 to 400 millse. even at that speed i'd be taking 30frames. – mvr950 Jun 4 '12 at 2:55
• Synchronization is easy, as it is just a brightness change caused by LED. You do not want to mess with shutter, delays, who knows what is inside software/firmware/driver stack. So simple independent TTL or 24V edge driving LED will illuminate scenery and let software choose what frames from what cameras belong to what time by simply recognizing illuminated frames in sequence. – user924 Jun 4 '12 at 3:25
• @ Rocket Surgeon yes that's a good choice. Also there's the possibility of pinpointing the location in cam2 frame B(that was pictured in cam1 frame A) by calculating Homography from static images. So you can also do that. – mvr950 Jun 4 '12 at 3:31

I would do this with IR LEDs. Check Vishay's web site for various LEDs.

The critical item you need to think is your FOV. (Field of View). From FOV and distance of your object to your camera, you must generate an approximate angle. Than go choose the LED that has the best fit. (The objective is not waste light) If there is no good fit, you can check out LEDIL web site for finding optical elements that would collimate the light into what you desire.

If you do this with IR, you can add a filter to your CAM and make only the reflectors show up since they will have an SNR at least 20-40dB above ambient objects. (Make sure that your webcam supports IR, they usually add a filter that blocks anything above 700nm. Almost all the security cams I know of doesn't include this filter so that is an option).

You also need to consider FPS, if your object is moving very fast, you will have trouble tracking your markers.

Final point, if you find a camera that allows you to control exposure time it is best, since auto control in the camera may not work well for your application.

• IR LED's are nice, but only if your camera will actually pick that up. While the sensor in most camera is sensitive to IR, many cameras have an IR blocking filter (of varying quality). – user3624 Jun 4 '12 at 1:51
• @DavidKessner this point was mentioned in the original respones. "Make sure that your webcam supports IR, they usually add a filter that blocks anything above 700nm. Almost all the security cams I know of doesn't include this filter so that is an option" – Ktc Jun 4 '12 at 2:23
• If you're extra-daring, you could pull a camera that you know has an IR filter on it and remove the IR filter from the sensor. See here: lifepixel.com/tutorials/infrared-diy-tutorials/canon-20d – Shamtam Jun 4 '12 at 2:44
• I was also thinking about cctv cameras. But the problem is low frame rate. Also I want to complete this project within budget. so money is an issue. Infrared led is a good option. Thanks for that. @Ktc Do you think ps3eye camera has filter too? – mvr950 Jun 4 '12 at 3:06
• @mvr950 No idea on the PS3Eye. But please be aware, your problem is probably exposure time, not frame rate. What is the motion you are trying to model? If you are doing gait analysis etc. 30fps would work well. If you have faster (sports) the 30fps may still work but you must have a strong light and short exposure time so that you avoid motion blur (smudge). – Ktc Jun 4 '12 at 7:37