A single-scan algorithm for connected components labelling in a traffic monitoring application

This paper presents a fast algorithm based on sequential local operations which aims at labelling connected components in binary images. While classical algorithms scan the image twice and utilize an equivalence table to store and resolve label redundancies, our method performs just a single scan, relying on the idea of labelling a whole blob at a time. In this way, we avoid label redundancies. As a consequence, the use of both equivalence tables and algorithms to resolve them becomes unnecessary. This leads our labelling algorithm to attain even more significant performances in the case of images characterized by blobs generating a large number of label equivalences. The proposed labelling algorithm has been successfully utilized in our visual surveillance system.