Guide to Cameras for Interactive Installations

Original Text by Blair Neal in 2013 with major update in 2020

To-do for 2020 Update:

Clean up language and links
Publish a version to Medium?
Update references to various camera technologies in all categories
Reference different connection protocols now available or more common (HDMI Capture, NDI, IP Cameras, GigE)
Webcam update technologies (Logitech Brio)
Other RGB Cameras (PTZ, 4K, DSLR, Broadcast)
Major depth camera update (Kinect Azure, Orbbec, Realsense, include Stimulant Depth camera shootout link, GigE Realsense
Thermal camera update (portable cams, USB, cheaper)
High Speed cameras (Edgertronic)
Others? (Leap Motion, GoPro, motion capture systems, things like Blacktrak(https://blacktrax.cast-soft.com) ), other things to consider as cameras or "observational tracking devices"
Older camera technologies (DVCam, RCA/Analog, "Analog" capture devices, etc)
Experimental
Maybe more general tips outside of sepcific camera technologies? Suggestions below from Elliot
Monochrome advantages
USB3 vs GigE
Lenses
Noise and Noise reduction
Machine learning touchup (upscaling, noise reduction, low light)
Using a smartphone camera (pros, cons, intgerated sensor data and intrinsics)
High end machine vision cams

Preface:

I published this guide in 2013 on Creative Applications and moved it to Github in 2016. I don't think the Kinect 2 was out at the time, and it has since come and gone. Camera technologies have obviously changed a lot in the 7 years since I published, so it is certainly time for a major update. The 2013 version is being kept in the repo for historical reference.

If your work involves interactive installations, there's a good chance you'll be working with a camera at some point. Choosing the right type of camera for your interactive installation is one of the most important technical choices you can make in the initial planning phases of your project. Making the incorrect choice can really impact how well your installation reacts to its visitors. This choice can also affect the installation's ability to perform robustly in a variety of environments. You can always correct for certain things in software, but the hardware setup can often be the first line of defense against undesired behavior.

Another important thing to keep in mind is that while many artistically geared software environments do work off the shelf with various cameras and input types, some more exotic cameras can be tricky to work with. Make sure you do your research before making a big purchase. If the tool doesn't exist to port it in, there are a few technologies that can route video between applications or even over the network. The most common tools are Syphon for macOS and Spout for Windows, but there are other options out there as well.

In the sections below, I will cover some of the most common camera types that are used for interactive installations. Before we dive in, it can be important to consider the following questions when considering which camera best suits your use case.

Questions to consider when in planning phases:

Where is it being set up? Does it need to be hidden?
Will it be outside and exposed to the elements or in the presence of direct sunlight?
What is the lighting control like in the space? Does it change throughout the day?
How precise does the interactivity need to be? Are you tracking large motions or small ones?
How big is the space you're trying to track people in? Where is the "active" area?
How will you run a cable from the camera to your computer? Will cable length be an issue?
Is your software compatible with your choice of camera?
Does the camera have settings that you can manually control or will it adjust itself automatically? If you need to control the settings automatically, does the camera allow for that?
Do you need a high quality live feed image that the visitor will see, or can you get by with low resolution for tracking purposes?
Is there a lot of movement in the space or in the background? How will you control for that?

These are the general camera options covered below:

The Basic RGB Webcam
other RGB Cameras
Infrared Cameras
Depth and Stereo Cameras
Thermal Cameras
Less common - high speed cameras, wireless cameras, motion capture, etc
Experimental Technologies

##1. The Basic Webcam - RGB

image source

Webcams are still great for basic applications and workspace testing, but are only occaisionally used for professional installations. Now that high quality cameras are built into almost every laptop and smartphone, the market for standalone webcams has largely stagnated since the mid 2010's. There are only a handful of decent options in this space and there doesn't seem to be a lot of choice or competition that will shake up the market anytime soon. A lot of the cameras are really made for Streamers and boast features for things like built-in AI face detection for autofocus (something that you may or may not even be able to use in your software). At this point, the Logitech Brio 4K is probably the best option and even that camera is almost 4 years old. Logitech is basically the only name brand in the game at this point, other than a random mix of Chinese manufacturers.

Connection types: USB 3.0 and USB C is most common, but there are still some USB 2.0 connections out there.

Max resolution range (typical): This value ranges a lot, and it depends on your application. There are only a handful of 4k webcams out there, but there are a lot more 720p60fps cameras on the market these days due to the adoption of USB 3.0. If you're working with an older camera, USB 2 often does not have enough bandwidth to send 30fps worth of 1080p without some artifacts. Some web cameras are also guilty of doing some software compression of the image before sending it through the hardware connection and this can add some image degradation.

Webcam Pros:

Cheap! - best widely available ones I've seen are around $80 bucks, but can get them for as low as $5 or $10 at this point
Easy to find
Reliable - leave them on for months (not always though!)
Full color image - they can see just about everything you can. Some can see some IR wavelengths. Rarely will you actually need the color image since most commonly used CV algorithms at the moment use monochrome images.
Widely supported by just about any software environment, and well understood.
They can see projection and content on screens. Essentially they see a less dynamic/contrasty version of what you see with your own eyes.

Webcam Cons:

They can see projection and content on screens.
Some brands will have really poor image quality, especially in low light.
Excessive image noise can severely impact tracking algorithms, especially in low light.
Rarely do you get to manually change the settings of the camera itself without some software investigation.
Typically a fixed lens (no zooming or manual focus), but manual ones do exist
Sensitive to changes in daylight/natural light
Requires fairly sophisticated computer vision algorithms to extract really meaningful information from these cameras - no skeleton tracking or easy depth information
USB cable lengths can be limiting if you need to be extremely far away from your processing computer. Plan for extenders/repeaters if going much over 30ft (10m) away
Latency can sometimes be an issue with these. Not too much, but enough to feel just a little bit behind real life. Analog is a little bit faster in regard to processing the image. Also most of them, aside from the PSEye tend to max out at 30fps, and 60fps is really recommended for a lower feeling of latency and high tracking accuracy.
Some cameras have weird methods of autofocus that can really screw up your imagery. I've used a few that were nice HD cams, but the slow hunting autofocus made them a deal breaker.
Try leaving your camera on for an extended period of time. I've seen issues with inexplicable strange coloring, and just non-responsiveness after being left on for a long time with certain brands.

Range of effectiveness: Varies by placement. Could watch an entire room of people, but with very little precision. Seems to work best in a range of about 1ft-50ft (.3-16m). If you're too far back, it can be difficult to get an accurate or meaningful reading because the people will be too small to the sensor in relation to other image noise. Can be fine if they are on a fairly plain background though.

Optimal environment for repeatability/reliability:

This is one of your only options on the list that is able to see the full color range, so you're sort of stuck with this or the pricier options in the next spot if you're doing any tracking that plans on using the colors in your environment.

Ideally a controlled room with good artificial lighting and no windows will ensure this camera will react the same at any hour of the day.

Most simple tracking methods with this camera will work best when the subject is on a plain solid background so it is easy for your software to pick out a person from other objects in the space. Background subtraction is suggested in a case like this.

Troublesome environments:

An environment with a lot of changing daylight or natural light can really impact this camera's ability to reliably and predictably track at all hours.

A room with none or very little light will pose a lot of issues for getting reliable readings from this camera.

Rapidly changing lights or flashing projection in a space can sometimes upset their calibration.

Webcam Further reading (some of these links are for older systems):

Check out the differences between a UVC cam and a non UVC cam to get an idea about if you'll be able to manually control the camera from within your software. Check out UVC camera control for macOS here Focal Length Calculator Spreadsheet of different current webcams and their capabilities for manual control on macOS Check out Macam for additional support for certain specialized webcam uses.

##2. Other RGB cameras

Image source

For this category, I'm thinking of things like HD/DV camcorders over Firewire, an HD-SDI or component capture card, or high end production cameras like the Red Epic or Arri Alexa (ok, maybe those are a little ridiculous). There are various reasons for wanting something in this category. and it usually comes down to needing really high quality and high resolution images at a very low latency. You may also be working with some unusual equipment, capturing input from another computer or image device that isn't a camera or have some other concerns about image input. There are still a ton of things you can do with analog cameras that you simply cannot do yet with all digital setups.

These cameras tend to have the same qualities/pros/cons as the standard webcam does. There are just a few important differences in certain use cases.

Beware of interlacing from analog capture devices. Tracking algorithms will not be a fan of that. Also be aware of anything involving NTSC versus PAL. You will be able to capture at a higher resolution and full frame rate, sometimes up to 60fps (or higher!) and 1080p and beyond, but that comes at a hefty price tag and will consume a large amount of computational resources. Some capture tools let you do full 4K resolution preview monitoring, but these occasionally need their own software pipes to get something that big into your processor. Check out this <a href=">Black Magic 4K Ultra Studio for an option for getting 4K input.

There are some fantastic options for getting HDMI/SDI input into laptops now via thunderbolt connections, like the Blackmagic Ultrastudio or Blackmagic Intensity. These capture devices won't always work if you're sending signal from another computer though, so check your video color formats first for compatibility.

Capturing with a DSLR all day would be great because you'll get the best image for the price range, but those cameras are not always designed to be left on for extended durations. DSLR's can easily suffer from heating issues or auto-shutoff if left on for too long. A more robust production camera or a solid DV Cam is more ideal for long running installations that require a really good image (and can't use a webcam for whatever reason), but make sure to do a lengthy test run before installing. Note: if tested and implemented with care, some DSLR's can perform just fine for long installation runs (one person has mentioned one running for 6 months at 7 hours a day with no issues).

DSLR's may need their SDK and some software massaging to work correctly in your environment, although some offer hit or miss HDMI output that would require a capture card. Canon's EDSDK for example has been ported to openFrameworks for easy access, and Magic Lantern Firmware.

Other high end cameras also enjoy the benefits of working with cables that were designed with long cable runs in mind. Webcams over really long (10m/30ft+) USB cables can cause jitter or some other funky consistency issues, so again...test before installing whenever possible. Production quality cameras can work over HD-SDI for fairly long runs (around 200-500ft depending on who you ask), but for the really far stuff, you're best either going over a network if you don't mind latency, or going over fiber with a converter box. Fiber will get you considerably further, more than 2500ft if you need it. You'll also need a capture card for some of these, occasionally requiring a tower instead of a laptop. Adding length in almost any case is going to add to the potential for problems, so if it is something going over 200 or more feet, try to check your whole chain first.

You're sometimes limited by the capture device and its support on your intended system. Some capture devices need special drivers or other magical mysticism to work within your intended environment, so be warned before going down this path...save your receipts.

Further Reading:

Frieder Weiss's writeup on using digital versus analog cameras and the latency issues involved - 2008)

##3. Infrared Cameras

These are really an essential tool for anyone working with interactive installations. Our visible light goes from 400-700nm, but certain cameras have the ability to see from 700-1000nm. This allows you to hide tracking lights (IR LED's or IR floodlights) and do other tracking of things not visible to the human eye.

The Playstation Eye is sort of the go-to for a lot of these applications (it's used in projects like the Eyewriter) because it does 60fps (or higher!), has a range of lenses you can buy, and it's pretty cheap for the quality. Read more about it here. It will need to be modified and requires a visible light filter to only get the IR bands. It is also a great option as a regular RGB cam since it has variable focus and you can buy extra lenses for it to get wide or narrow angles.

For the next level up, the other common one is the Point Grey Firefly. One tip for this one is to go ahead and buy only the monochrome version of this camera, as the color version has the additional color filter layers that can impact image quality and response. I stick with the firewire version of this camera for the bandwidth, but that can limit you on cable length (you can really only get firewire out about 10-15ft without using a repeater or amplifer, so keep that in mind when building equipment lists). The Firefly will also take a range of lenses that give you an insanely large aperture of around 1.4-1.2f and will be able to give you a much brighter image than a standard webcam. You WILL need a visible light filter for the Firefly to only get IR bands though. See Golan Levin's article on choosing the right IR filter for your application. In addition to the Point Grey, you can check out these cameras from Imaging Source.

When choosing these higher end cameras, also make note of their sensor size. A larger sensor will potentially increase your output resolution and decrease sensor noise, but you are left with a smaller range of lens choices. The happy medium seems to be a 1/3" sensor with a CS mount for the widest range of options.

Connection Types: USB 2 (Playstation Eye or another webcam - modified), USB 3 (Point Grey Flea 3), Kinect's IR cam, Composite/Firewire (old Sonycam's with night vision)

Resolution Range: You can generally get the same resolutions in IR that you can in regular webcams. There are plenty of tutorials out there for modifying standard webcams to only process IR light.

**Range of effectiveness:**Generally the same as a webcam (1ft-50ft or .3-16m). Some IR cams like the Point Grey have a wider range of security camera lenses for various zoom ranges if you need to do tracking from far away.

Infrared Pros:

They cannot see images that are projected or on a screen (some can see content on screens...depends on screen type). IR and projection are a really nice interactive match, but you tend to need a dark space to pull these off. Are you also tired of being relegated to a dark windowless room with these tools?
Can be used in dark spaces when used with IR emitters.
Since they can see what the human eye can't, they are most useful for hiding certain elements like lights used for tracking points or illuminating a space.
Tracking a point of IR light is an incredibly effective and robust tracking method because you can have a bright point of light that is invisible to the human eye.
Good for tracking a large area like a stage flooded with IR.

Infrared Cons:

Monochrome
They require a healthy source of infrared light. Don't assume the lights inside will provide the type of light these cameras need to see effectively. You may need IR emitters to properly illuminate the space. People wont be able to see the extra light, but the camera will.
Sensitive to sunlight and certain stage lights. Some stage lights will be bright enough to throw off tracking on an IR camera because the range of the light and camera overlap enough. There may be problems that the camera can see that you might not have seen with your eyes, so it can be useful to know the complete light profile of a room or to just go there with a camera beforehand.
Certain clothes and materials look different in IR than in visible light. Can have weird effects sometimes.

Optimal environment for round the clock reliability

Room with a steady amount of IR light, either artificial or coming from a point source. Undesired sunlight can blow out an IR cam if the aperture isn't set right or if it doesn't have an automatic adjustment.

Troublesome environments

A dark room with insufficient IR light. A lot of normal light bulbs will still show up in IR even if they don't put out enough light to illuminate the IR spectrum, so keep an eye out for those in the background as pesky tracking distractions.

Further Reading for IR cams

Frieder Weiss's writeup on IR cameras and equipment -2008

Other IR camera suggestions: Cheap Infrared CCTV cams work well, but you'll need a capture device Sony CCTV cam good in low light (needs light filter) The Sony M183 or M383 are classic low latency cams used for IR sensing, but are now discontinued, but check eBay.

##4. Depth Cameras

(image source)

In a way, these are an offshoot of the IR camera category but they involve some form of infrared light projection that gets reassembled as a depth image using a variety of techniques like time-of-flight (TOF) or structured light. These two detection methods are pretty different and have their own pros and cons, in terms of potential speed: read about their differences here.

The information that comes back in a depth image is much richer than the usual "2D" image captured by the cameras above, and it gives you the ability to more easily and robustly detect things like hands and fingers. Tracking those features would be difficult and require very controlled conditions with just a monochrome 2D image. The depth information makes it easier for you to detect the difference between a person and the background, and therefore assign more meaningful information to the human trying to interact. This is why it is easier for the Kinect to give you a person's skeletal information, something that would be difficult for just a regular 2D IR image.

There are also other methods of obtaining the same information with a high framerate webcam and a projector (structured light). Another method called LIDAR scanning uses lasers and a method of measuring the reflected light to construct a high resolution depth image. LIDAR is usually more effective than other methods at measuring an incredibly large area or the front of a building (for quickly getting a 3D model of the facade for a projection mapping surface).

Connection types: Primarily USB2 (Kinect, Asus Xtion, Panasonic D-Imager, Intel/Creative Time of Flight cams), other interfaces in more research oriented/industrial cameras.

Resolution Range: This mostly tends to top out around 320x240 @30fps for the more consumer level cameras, mostly for TOF. TOF cameras use different sensors than standard CMOS cams which impacts their speed and resolution. The Kinect is faster and higher resolution because they just use a standard IR cam. Check the individual camera for it's maximum resolution.

Range of effectiveness: Typically less than 10-12ft ( 3-4m). Some cameras like the Intel/Creative TOF camera are really only designed for desk use and give you about 1m of effective distance, if that. Other TOF cameras like the D-Imager can give you up to 10m of range though. Makes certain sensors not very robust for use on large stages. Requires multiple sensors to cover a large area and thus complicates the CV processing chain and hardware setup. LIDAR can do a large area, but not necessarily in real time.

Depth Camera Pros:

Depth images give you the option of really amazing background subtraction because you can totally ignore everything beyond about 10ft (with a Kinect).
Processed depth images give you a lot of great tracking information like hand skeleton, body skeleton, shape normals. You can build 3D models of rooms or objects. You can do fairly high quality body and facial motion capture for use in 3D rendering programs.
Isn't affected by light from a projector. A lot of the same good qualities as standard IR, but they should only be used if you need depth information...they are not preferred over regular IR. Again, do your best to choose right tool for the job.

Depth Camera Cons:

Because of the use of IR, these cameras will also be sensitive to changes in sunlight.
Do not rely on depth cameras if you need a sharp, clean outline of your subject. You will get jagged artifacts that will need to be blurred or processed to get smooth lines.
Structured light cameras aren't able to work at an incredibly close range (ie right in front of them). Time of flight cameras are a little more robust in this regard, and are better if you need to use them really close up.
Certain surfaces will interfere with the ability to produce a good depth image. Shiny surfaces will effectively be invisible or cause strange reflection artifacts in your image.
Using multiple systems together occasionally requires special consideration. Overlapping structured light from the Kinect (if they are right next to each other) can cause double images to appear due to the extra light grid. If the Kinects are more perpendicular to each other, this negative effect is less likely. You can also use this trick: attach a vibrating motor to one Kinect to make the other sensor's dots blur, and the vibrating sensor's own dots will stay in focus.
Fairly limited effective range.

Depth Image Examples:

LIDAR/Structured light was used to get the massive scans of the spaces in this Radiohead video for House of Cards:

Further reading:

If you're just looking for stereo vision capture that can infer depth info, check out the Readings on structured light and how it works

-- ##5. Thermal Cameras

(image source)

http://en.wikipedia.org/wiki/Thermographic_camera

These cameras are comparatively rare to see in use in interactive installations because they are still prohibitively expensive compared to most other ones. They aren't totally unobtainable ( roughly $1000-20000) but that higher price tag makes them a little less desirable for early exploration on projects. It's a shame because these cameras offer a lot of abilities that just aren't possible with the other kinds of cameras.

There are various types of cameras to look at in this class, mostly pertaining to which part of the IR spectrum you're trying to see. You have the option of Long Wave IR, Mid Wave IR, and Short Wave IR. For thermal imaging, you'll mostly want to work with Long Wave IR, in the 7000-14000nm range.

I have not personally used one of these cameras yet, but they have some properties that would be really amazing in the tool belt of people making interactive installations.

Check out this guy doing some random demos with a thermal camera:

Connection types: Most are made to be integrated into existing systems and either have proprietary connections or just output composite video. Some cameras communicate X/Y position of blobs.

Resolution range: Some are very low resolution (the Thermitrack is 16 x 16px) and some are close to a VGA range, but don't expect to find HD thermal for cheap. You also get a variety of frame rates and contrast ranges.

Thermal Camera Pros:

Normal visible light doesn't have much of an effect on a thermal camera.
Good for tracking a large area like a stage.
Gives you the ability to more definitely identify people because of their heat signature...whereas other objects and materials may not show up at all if they are warmer.
Fairly robust for daytime to night time interaction because people will be the same temperature and appear in the proper dynamic range.
Give you the ability to track invisible phenomena like hot air from breath, or residual heat from something like a hand leaving a warm mark on a surface, or a cold blast of water hitting a warm surface.
Can see through certain materials and walls.
Thermal cameras can see through certain kinds of clothing.

Thermal Camera Cons:

As these are occasionally considered military grade equipment, there may be export restrictions, so be wary when planning on traveling abroad.
Expensive
Difficult to integrate - require either custom electronics or capture hardware
Thermal cameras can see through certain kinds of clothing.
Not all lights are invisible to thermal cameras...if it's producing heat or radiating it, the camera will see it.
Thermal imaging sometimes results in ghosting of movement due to the sensor method.
They are unable to see through windows/glass because the range of radiation ends up being reflected before it transmits through the glass to the camera. See here for a more involved explanation of why.
Certain hot materials may result in unforeseen difficulties with using thermal imaging in certain environments. Requires a different type of thinking in order to anticipate things that will be overly hot or cold in the space of the installation.

Further reading:

People counters (wiki)"

Thermitrack

##6. Experimental technologies

These are cameras that haven't really reached the public market yet, or just haven't had the chance to be explored for interactive installations. This category can also include less conventional means of getting an image or tracking objects, or other tools that can be effected by visible light in the space (for example, the Leap Motion controller can be thrown off by excessive amounts of IR light). There are also light field cameras, like the Lytro, that are able to get a different kind of information for shooting a scene than a normal CMOS detector can get. Some of these technologies are either brand new or too expensive or not accessible by most so they don't get a lot of use in interactive installations (yet) but it's only a matter of time for several of them.

Lytro example:

Notes:

Huge thanks to Kyle McDonald, Theo Watson, Zach Lieberman and Golan Levin for some great additional tips to help round out some of the suggestions here.

The above information is really from my own experience of using various camera technologies in interactive settings. As usual, try to do your own research first and use this as a guide of someone else's experiences. Some cameras I've used more (webcam), some less or not at all (thermal), so take some of my suggestions with a grain of salt although some pointers I've absorbed from some of the experts in the field.

Also, it would be good to go hit up wikipedia or other more accredited sources for the technical information on the visible light spectrum versus near and far infrared if you need a refresher, I have left most of those details out. Really understanding the science behind how the different cameras work can expand how well you actually use the camera in practice.

jermad/Guide_To_Cameras_Interactive_Installations

Guide to Cameras for Interactive Installations

Original Text by Blair Neal in 2013 with major update in 2020

Image source