Computer Engineering Paper Alberto Broggi and Andrea Gandini, Parallel Image Clustering: A Real-Time Application for a Special-Purpose Architecture, In Kjell Arild Hogda, Bjorn Braathen, and Karsten Heia, editors, Proceedings 8th SCIA - Scandinavian Conference on Image Analysis, volume 1, pages 297-304, Tromsoe, Norway, May 25-28 1993. IAPR - International Association for Pattern Recognition, NOBIM - Norwegian Society for Image Processing and Pattern Recognition,
Download the paper in compressed PostScript format
.

ABSTRACT

The main topic of this paper is the early visual processing of images grabbed by a camera placed on a vehicle, a car, moving in a real road traffic scenario. More precisely, the paper presents a parallel algorithm for real-time segmentation of images. Due to the constraints imposed by this kind of applications (real-time response), a special-purpose massively parallel SIMD architecture, PAPRICA, has been developed; the segmentation algorithm has been conceived and designed to be implemented on such an architecture, and on parallel mesh-connected machines in general. This work is just an example of the application of PAPRICA architecture to real-time problems, where the reduction of the processing speed and the performance analysis play a fundamental role in the algorithm design stage. In fact, as often happens in industrial applications, the solution to real problems which require high performances needs to be simple and well-tuned. Since PAPRICA hardware is now under testing, the first version of this algorithm has been implemented on its simulator running on a Connection Machine to be studied and tuned, in order to achieve the real-time version for PAPRICA system.


For more information send email to broggi@CE.UniPR.IT