Cameras

To get an image into a computer, you either need a video camera and a digitizer, a digital camera, or hard copy and a scanner. The department has one scanner, hooked to a MAC in the 7th floor software lab, which could be useful to you occasionally. Most of your image taking, however, will with video cameras, of which we have several.

In the vision lab, there are two or three pulnix color cameras that put out rgb and composite video. The rgb connectors are on a separate breakout box that is attached to the power supply. The composite can be fed into one of the composite ports of the BT848 digitizers on the computational sensors. The rgb can be fed into the KTV digitzers in Brain and Aorta. When not in use, these cameras are stored on the lefthand modular shelf unit. These cameras use C-mount lenses of which several, of different focal lengths, are kept in the gray cabinet against the wall.
There are three monochrome C-mount cameras with self-contained power supplies that plug into a wall outlet or extension cord. These output a monchrome signal that can be fed into the composite channel of the bt848 digitizers, or the green channel of the KTV digitizers. These cameras have mostly been used in conjunction with a black velvet turntable and the robot Hillary to acquire multiple images of single objects from various views for use in training and testing recognition systems, but there might be other uses. Talk to me if you want to use this system, since the Puma robots are the only pieces of equipment we have that are not compliant with the first and third laws of robotics: that is, they can hurt you or other people, and they can damage themselves if given the wrong command. (They may also not do what you tell them to, or at least what you think you told them to, but that is shared by other computational devices).
There are a number of Sony EVI pan-tilt cameras, and several new Canon pan-tilt-zoom ceiling-mount cameras. Both put out color composite video, and s-video (YC) signals, and are generally used in conjunction with the computational sensor platforms. The image quality of the Sonys seems a bit better than that of the Canons (I think the chip may be larger in area), and the autofocus works better, but the Canons are much smaller, lighter, have a greater zoom, pan and tilt, and mount cleanly on the ceiling.
The computational sensor platforms are dual-processor, Intel (Pentium/Xeon) boxes running Linux. There is an assortment of these, representing various purchases experiments. The fastest currently have 2.0GHz Xeon processors. (With a set of 3.0GHz boxes to be fielded in the near future). They have a load of memory (up to 1GB) and a pile of local RAID storage (up to 280GB per box). They each have two BT848 digitizers, so they can handle dual cameras. You can put either the composite, or the S-video signals into these digitizers, but the S-video is better quality.
You can get monochrome into the KTV digitzers by running the Y signal from a s-video splitter into the green channel. To get color, you have to run the composite signal through a composite-to-rgb converter (there is one in the rack in the vision lab) and then into the KTV digitizer. A nuisance.
The Pan-tile-zoom cameras have 12 to 1 zoom (more on the Canon model), autofocus and autoexposure if you want it, and pan and tilt control as well. You can control the cameras with a remote (like for a TV) or from a computer over a serial line. There is a C++ library for doing this in with my other libraries, /u/nelson/programs/lib/Solaris/libcamera.a; source and demo programs in /u/nelson/programs/src/robot/pt_camera/[lib, bin]. The library supports both the Sony and Canon cameras. The C++ library does not access every possible function of these cameras. If you want to know more about the Sony camera you can take a look at the documentation of the serial control interface (this is a pdf document).
Currently, there are about 10 of the Sony cameras mounted in the vision lab, and some other loose on tripods or on the camera shelves. There is one Canon camera on the ceiling, and a lot of others in boxes, soon to be opened (10/2003)
You can use these cameras to remotely monitor what is going on in the vision lab (or anywhere else one of them happens to be). To do this, first make sure the camera is turned on. Then bring up a remote X-window on the machine that has the camera and digitizer hooked to it. Make sure your DISPLAY environment variable is set to the machine you are on. Run the program bt_showlive in my bin (~nelson/bin/PCLinux) if the camera is attached to one of the Linux boxes with BT848 digitizers, or the program ktv_showlive (in ~nelson/bin/Solaris) if it is attached to a Sun machine with a KTV digitizer. This should bring up a live image on your screen, transmitted over the ethernet. By default, the image is at half resolution and in monochrome, but you can make it bigger or smaller, and run it in color (assuming the digitizer is digitizing color) with various options (use -h to get a list). Then run the program camera_demo also in my bin/[PCLInux|Solaris]. This will let you control the camera (pan, tilt, zoom etc.) from the keyboard. Run too many of these, and you will bog down the local network. It is also considered poor etiquette to harass people by spying on them or their terminals with these cameras.
On our old robot head, there are a couple of miniature color cameras. These put out rgb and composite video, and take miniature lenses, a selection of which is also kept in the gray cabinet. There is code somewhere for running the head, or at least the motors. Raj Rao was our last expert on it and he's gone now, but I can probably dig up something if anyone is interested. The motor source is in /usr/src/vision/[motor, motors] for a start.
*** This was sent to Rod Grupen at UMass ***
We also have a newer robot head, which has a couple of very high quality monochrome cameras on it, equipped with lenses with controllable iris, focus, and zoom. The problem is, the cameras are pretty useless without the software to control the head and the lenses, and the control interface for this beast was so unwieldy that we never really got it working, and what little software we had has long since rotted. This head is currently mothballed in the gray cabinet in the vision lab. However, we now have one of the designers Brian Madden, previously working for Kyros, still over in the Dermatolgy lab, so if anyone is inclined to get this going again...
We have two USB cameras that have smaller chips, and thus lower quality images than the Sony cameras (though they are supposed to be better quality than your average web-cam). We do not currently have a well developed software interface for using these cameras, especially from within programs. It is claimed that the program "coriander" can be used to get pictures from thes cameras. A version is sitting in Brandons directory ~sanders/Linux/bin/coriander, and it is available on the web. It seems buggy to me. The various compression modes don't seem to produce output commensurate with their nominal quality, and I have managed to crash Linux by trying to use both coriander and the BT848 digitizers at the same time.
There are also a couple of consumer video cameras - the kind that people record their kid's birthday parties on. These output s-video (YC) and composite video, have automatic exposure and focus, will stabilize images against your jittering hand, and have nice integral zoom lenses, and give you a tape into the bargain.
Kyros Kutulakos has built a lab over in the med center Dermatology department with a dozen or so high quality cameras with accurate calibration equipment. See Brian Madden if you are interested in well calibrated, multi-camera setups.
Finally, there is a digital camera in the lab. You can read the manual, take few pictures, plug it in to the USB port on one of the computational sensors, and if fortune is smiling, the images will show up as a directory from which you can transfer them to the department file system. The format and compression is controlled by the camera, so you may need to read the manual which should be in the vision lab somewhere. I have not used this camera, so this description may be off, but people have obtained pictures with this camera. The quality and resolution is far better than that produced by the Sonys (or any of our video cameras), so if you want a good color pic, this is the way to go.

Back to vision course main page