MEGA-D Megapixel Digital Stereo Head

Image Gallery This image gallery shows some of the output from the STH-MD1(-C) in different formats.  One of the most interesting properties of the MEGA-D is the ability to do electronic vergence.  Instead of mechanically pointing the two imagers towards a close object, the imagers are sampled at different horizontal offsets to center the object.  Mechanical vergence is problematic for stereo calibration; electronic vergence preserves calibration. NOTE: All these images are compressed JPEG, so there is some degradation of the quality from the original.

   Sample Outdoor Images

Sample color images from the MEGA-D, at 640x480 resolution.  We took these pictures at SRI as part of the DARPA Perceptor project, for automatic guidance of an outdoor vehicle.  Click on the image to see a larger format.

	A 640x480 color image of an outdoor scene at SRI.  This is the left image of a stereo pair.  The image has been rectified to correct for lens distortion and imperfections in the stereo rig.
	This screen shot of the Small Vision System shows the disparity image obtained from the stereo pair.  Brighter green indicates a larger disparity, hence closer to the cameras.
	Here is a 3D reconstruction of the scene, using the OpenGL interface of the SVS system.  The original camera position is shown by the three axis lines.  The red line indicates the camera optic axis.

    Color Images

These are images from the STH-MD1-C. The PixelCam CMOS color imagers provide excellent color resolution, because the megapixel format means you can get true colors at each pixel of a 640x480 image without interpolation! Click on the image to see a larger format.

Full-frame at different resolutions

1280x960	640x480, binned	640x480, decimated	320x240, bin+decimate

These images all show the same scene, taken with the MEGA-D with an 8.5 mm lens (60 deg FOV). Click on the thumbnail to get a full-size image. These images illustrate the quality and resolution obtainable with the PixelCam imagers.

• Excellent resolution, with extremely fine detail at the highest resolution
• High dynamic range
• Low noise. Binning averages 4 adjacent pixels, further reducing noise.
• No smearing, streaking or blooming under high intensity lights. Note especially that saturated parts of the image (flourescent lights, phone highlights) do not smear onto adjacent pixels.

Subwindowing

An arbitrary subwindow within the full frame can be returned, allowing for faster data transmission and realtime panning. Click on the thumbnails below to see a full-size image. The 640x480 subwindow is taken from the middle of the original 1280x960 image. The other three images are 320x240 subwindows taken at various points in the 1280x960 image.

Vergence

One of the great things about subwindowing a large-format image is that you can perform electronic vergence. When an object is close to the cameras, and the cameras look "straight ahead," here's the image you get (320x240 subwindow):

There isn't a lot of overlap between the two images. The cameras need to be pointed to the subject, or "verged," in order to capture more of a common image. Here's the result:

Mechanical verging is difficult, requiring motors, control loops, and complicated design. It also changes the calibration of the cameras, so that they must be re-calibrated when the vergence is changed.

Electronic vergence is much simpler. The selected right and left subwindows are moved from a common offset, so that the right image is further right than it would normally be, and the left image is further left. This effectively forms a verged image pair, as above. Because the vergence is done on the image plane, no re-calibration is necessary, and the vergence can be done in real time. The Mega-D supports real time electronic vergence.