User`s manual

STOC USER’S MANUAL ©2005 VIDERE DESIGN
1 Introduction
The STOC is a new type of device. It functions as a complete vision
processor, combining stereo cameras and an embedded processor to
produce 3D stereo information. The processor was developed at SRI
International’s Artificial Intelligence Lab, for use in real-time vision and
robotics projects. Because the embedded processor has been designed for
just this task, it is more capable than the fastest PC, achieving over 40 Gops
(40 billion operations per second). And it does this at a fraction of the
power – just 0.8 watts for the processor, and 2.4 watts for the whole device.
The result is an output stream of 3D information, at a resolution of
640x480, at 30 frames per second.
The stereo cameras are 640x480 (VGA), progressive scan CMOS imagers
mounted in a rigid body. They have a global shutter – all pixels are
exposed at exactly the same time. This makes the STOC suitable for
environments with fast movement, such as on an outdoor vehicle. The
imagers have excellent dynamic range, sensitivity, anti-blooming, and noise
characteristics. They are fully controllable: the user can set exposure, gain,
decimation, etc.
The STOC uses standard miniature lenses, or standard CS-mount lenses, for
user-changeable optics. Wide-angle to telephoto options are available,
depending on the application. The device comes fully calibrated, ready to
produce 3D information out of the box. A simple, in-field calibration
procedure is available if the lenses are changed.
The STOC has an IEEE 1394 (FireWire) interface for output and power. It
connects to any 6-pin FW port, and delivers real-time 3D information using
the DCAM 1.30 standard.
SRI’s Small Vision System (SVS) software has an interface to the STOC,
and is included with each stereo head. The SVS software includes software
drivers for the STOC for MS Windows 98SE/ME/2000/XP, and for Linux
2.4.and 2.6 kernels.
1.1 Characteristics of stereo system
• On-device stereo processing
Xilinx Spartan 3 – 1000 processor, 88 MHz
512 KB SRAM
• Stereo Processing
640x480, 64 disparities @ 30 Hz
Subpixel interpolation of disparity to 1/16 pixel
15x15 correlation window
9x9 Laplacian of Gaussian preprocessing kernel
Post-processing uniqueness check
9 cm baseline
• Fully calibrated for lens distortion at the factory
Recalibration in the field for change of lenses
• Firmware upgrades via IEEE1394 connection
• Power: 0.84 W for stereo processing, 2.4 W total
1.2 Characteristics of imagers
• Micron MT9V022 CMOS imagers
Global shutter
Simultaneous exposure and readout – true 30 fps
640x480 maximum image size
1/3” format
High sensitivity, low noise
Low pixel cross-talk
• Fully synchronized stereo – left and right pixels are
interleaved in the video stream
• Monochrome or Bayer Color
• High frame rates – 30 Hz for 640x480
• Extensive control of video parameters
Automatic or manual control of exposure and gain
4