SPECIALIZED PARALLEL STRUCTURES FOR VLSI IMPLEMENTATION OF THE HOUGH TRANSFORM FOR ARBITRARY SHAPE DETECTION (COMPUTER VISION, HARDWARE IMPLEMENTATION)

Scope and Method of Study. The purpose of this study is to design inexpensive and fast special purpose functional units for the Hough transform designed for arbitrary shape detection. Existing Hough transformations are actually implemented and tested with images containing industrial parts in order to analyze their performance on industrial parts recognition. These algorithms are also analyzed for suitability of parallel implementation. Based on these analyses, an existing Hough transform was selected and reformulated for efficient parallel implementation. Two parallel structures--a mesh connected structure and a linearly connected structure--are designed for the reformulated Hough transform. In addition, determining the orientation of an object by using the Hough transform is proposed to take advantage of these structures.; Findings and Conclusion. The performance analysis of the Hough transform on industrial part recognition shows that the algorithm has a great potential for industrial computer vision systems. The functional simulations of both proposed parallel structures shows that the Hough transform can be implemented in inexpensive, compact, and fast VLSI devices. These structures also promise a single chip Hough transformer for reasonable size image (256 by 256) within a few years. These inexpensive and compact transformers will boost the capability of current computer vision systems, which suffer from a lack of computation power for high level computer vision algorithms.; The test results of the algorithm proposed for orientation determination indicate that this can be a powerful algorithm for industrial computer vision systems if it is computed by the parallel structures proposed for the Hough transform.