Feature Extraction with Hough Seeded Region Growing as Data Compression for Distributed Computing

Abstract

The field of environmental perception is an important task to allow autonomous vehicle to navigate safely in crowded urban environments. Since the sensory technology is established for several years the main task lies in the data analysis. Especially imaging sensors collect a huge amount of data that contains a high redundancy. Due to this, the generation of an of a environmental picture with imagine sensors is challenging an requires sufficient computational Power. A possible solution is to distribute the necessary calculation on several smaller computational units, where each of them analyses only a part of the data. From this follows a higher bandwidth demand to the communication system since the data must be transmitted between the computational units. In this work a Region Growing Algorithm is combines with a Randomized Hough transformation to extract planes from a 3D-point cloud. Additionally the software can reduce the amount of data that must be transmitted in distributed computing environments since redundancy is removed. This reduction is done with Grahams scan algorithm that generates a Polygon describing the extracted Plane.