A Novel Clustering-based Algorithm for Curve Detection and Its Application to Passenger Recognition
Hanxi Li, Hong Yuan Zheng, Yang Wang · 2007
A novel chain code and a new clustering-based algorithm are proposed to detect the curve in the binary image. The novel chain code, which is termed angle chain code (ACC), is more proficient and precise in describing the local-shape of the edge than classic chain codes. Thanks to the ACC and the extraction of geometries, the clustering-based algorithm can detect the mathematic models of contours efficiently. Compared with the standard Hough transform (SHT), our algorithm is much faster (by over 690%) and requiring much less memory space. Furthermore, it inherits the high robustness form classic clustering-based approaches. In some situations, it is even more accurate. We implement the novel algorithm in an embedded system to estimate passengers flow on the bus. The detection rate is over 95% which indicates it is quite suitable for real-time application.