Most dense stereo correspondence algorithms start by establishing discrete pixel matches and later refine these matches to sub-pixel precision. Traditional sub-pixel refinement methods attempt to determine the precise location of points, in the secondary image, that correspond to discrete positions in the reference image. We show that this strategy can lead to a systematic bias associated with the violation of the general symmetry of matching cost functions. This bias produces random or coherent noise in the final reconstruction, but can be avoided by refining both image coordinates simultaneously, in a symmetric way. We demonstrate that the symmetric sub-pixel refinement strategy results in more accurate correspondences by avoiding bias while preserving detail. 1.

We present a new unified approach to FFT based image registration. Prior works divided the registration process into two steps: the first was based on phase correlation (PC) which provides pixel accurate registration [7], while the second step provides subpixel registration accuracy [2, 5]. By extending the PC method we derive a FFT based image registration algorithm which is able to estimate large translations with subpixel accuracy. The algorithm's properties resemble those of the Gradient Methods [6] while outperforming them by exhibiting supirior convergence range. Furthermore, the algorithm is robust when registering severely noisy images.

Abstract: A simple, flexible and programmable CMOS camera is presented in this paper. This camera is primary designed for precise single-line dimension and position measurements with subpixel resolution, but allows also a monochrome image acquisition that helps in camera positioning. Information from adjacent lines is also used to improve the single-line resolution. Camera is designed as a stand-alone device and uses the USB connection only as a remote terminal in the setup phase.

Development of hybrid MEMS devices has demonstrated a need for automatic microassembly strategies. Visual servoing techniques have shown great promise as a control strategy capable of sub-micron precision while compensating for many of the problems that exist in the micro domain, including thermal expansion of assembly devices and imprecisely modeled and calibrated sensors and actuators. This project develops rules for micropart design to aid in device assemblability with visual servoing techniques by ensuring that the microparts can be easily tracked and controlled using vision feedback. A criterion is presented that estimates part trackability based on the visual appearance of the part. This criterion is then used to microfabricate features to improve part trackability and hence, the assemblability of the device. The criterion considers the feature appearance when the part lies out of the optical system's depth-of-field. A Fourier optics based approach is used to simulate the visual appearance of microparts represented by CAD models using high resolution optical systems. This simulation is used to automatically design microfabricated features on microparts. These features are used to estimate the tracking accuracy of the MEMS parts to subpixel levels using interpolation techniques in optical flow based tracking. This allows MEMS parts to be assembled with precisions on the order of 20 nm with high magnification lens, using visual servoing strategies. The curvature of the SSD surface is used to predict the tracking accuracy of the feature designed irrespective of defocus level and settings of the lens. Results demonstrating the capabilities of design-for-microassembly rules using visual servoing microassembly strategies are presented.

It is sometimes necessary to measure the motion of an machine part which both rotates about and translates along a common axis. We call this compound rotary-linear motion or cylindrical motion. Unlike simple rotary and linear motion, sensors for cylindrical motion are not commonly available. We describe three novel methods for measuring cylindrical motion in the context of a precision machine tool: the tiltedmirror interferometer (theory and implementation) , the helicoid-mirror interferometer (theory only), and the machine-vision technique of absolute registration (theory only). The tilted mirror interferometer uses laser interferometry to sense the position of a mirror that is mounted to the end of the moving part at a slight tilt with respect to the normal plane of the axis of motion. We use interferometer optics which measure linear displacement, and we mount three of these optics on the fixed part of the machine so that their beams reflect against the tilted mirror.

In this dissertation we discuss a variety of techniques that advance the state of the art in the field of 3D shape acquisition from real world objects. The research was done in collaboration with