The thesis concerns binary vision for non-contact inspection and robotics applications in flexible manufacturing. Two-dimensional silhouettes of three-dimensional objects are processed to measure a number of features including area, perimeter, circularity, maximum radius, moments, and number of holes. This information is then used to identify and locate randomly positioned objects to provide visual feedback for part acquisition by an industrial robot or inspection tasks that might include checking dimensional tolerances, verification of hole placement, and the like.

Silhouettes are encoded in the system as a linked list of "vertex points" representing changes of direction on the contour. An algorithm and data structures for extraction of vertex points from a raster-scanned binary image have been developed. The method operates sequentially and no restrictions are imposed on the number or topology of silhouettes in each frame. Several existing contour tracing algorithms are reviewed;​ the new algorithm is shown to offer considerable improvement in execution time at the expense of a small increase in memory. It is also demonstrated that vertex point approximations use significantly fewer points than run-length segment representations.

System implementation is based on a 232 × 240 CID camera, an 8086/8087 single-board microcomputer, and two custom-built boards. The architecture features a multiple bus configuration designed for high speed, parallel operation of dedicated modules:​ grey-level histogram generation and binary image acquisition are at video-rates. The complete software is in EPROM, and includes feature extraction for recognition and location of silhouettes encoded as vertex point lists, a data base for prototype objects and a comprehensive command set.

The software also includes communication with a PUMA 600 robot via a serial interface as well as extensions to its VAL language;​ in this way the robot accesses visual information under program control. An example is given of parts sorting using the extended instruction set.

The results of a large number of statistical measurements are used to establish overall system performance.