RAIL

Introduction

RAIL (Road Recognition from Aerial Images using Inductive Learning) is a road recognition system under development at the Artificial Intelligence Laboratory, UNSW. It is a semi-automatic, multi-level, adaptive and trainable edge-based road recognition system intended to demonstrate the use of various Artificial Intelligence approcahes in this area.

Starting with edges, complex structures are built from simpler ones in multiple stages, beginning with image pre-processing and edge extraction. The road detection is split into four levels, covering road segment detection; road segment linking; intersection detection; and joining roads to intersections. A combinations of different different machine learning techniques can be applied at different levels. Segmentation and classification of features based on inductive machine learning and cluster analysis are key components within RAIL.

RAIL is a joint venture between the School of Computer Science & Engineering and the School of Surveying & Spatial Information Systems. It is funded by the Australian Research Council grant and industry.

Reports


06/02	New results for Inductive Clustering Improved evaluation Level 2 results as well as the rerun of level 1 on new evaluation Results on new images (morpeth) Doc file Image files PNG files C4.5 files

04/02	Summary paper submitted to ICSML Workshop 2002.

03/02	inductive clustering paper submitted to PVC'02. evaluation paper submitted to PRRS02.

02/02	A review of level one attributes Improved KNN

02/02 (Gary)	Image Attributes Implementing Level 2 Road Reference Model Format

01/02	High level RAIL models Inductive clustering experimental models

12/02 (Gary)	Edge-Based Attributes for Roads

11/01	Review of RAIL and its attributes for Level 1 & 2 kNN clustering algorithm and results Evaluation techniques and results are presented Rules generated by C4.5 (Updated Jan 02) Future direction

08/01	Implementation details of kNN

Publications

Chen, A., Donovan, G., Sowmya, A., Recent Progress on RAIL: Automating Clustering and Comparison of Different Road Classifcation Techniques in High Resolution Remotely Sensed Imagery. In: The Nineteenth International Conference on Machine Learning 2002. Sydney, Australia.

Chen, A., Donovan, G., Sowmya, A., A Comparative Evaluation of Different Road Classification Techniques in High Resolution Remotely Sensed Imagery In: The 2nd International workshop on Pattern Recognition for Remote Sensing 2002. Niagara Falls, Canada.

Chen, A., Donovan, G., Sowmya, A., Trinder, J., Inductive Clustering: Automating Low-level Segmentation In High Resolution Images. In: Photogrammetric Computer Vision ISPRS Commission III Symposium 2002. Austria.

Teoh, C.Y., Sowmya, A., 2000, "Junction Extraction from High Resolution Images by Composite Learning", In: The Internaional Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Amsterdam, Netherlands, Vol. XXXII, Part B3, pp882-888.

Teoh, C.Y., Sowmya, A., Bandyopadhyay S., 2000, "Road Extraction from high resolution images by composite learning", In:Proceedings of Internaional Conf. Advances in Intelligent Systems: Theory and Applications, Amsterdam, Netherlands, pp308-313.

Sowmya, A., Singh, S., 1999, "RAIL: Extracting road segments from aerial images using machine learning", In:Proc. ICML 99 Workshop on Learning in Vision, pp. 8-19.

Singh, S., Sowmya, A., 1998, " RAIL: Road Recognition from Aerial Images Using Inductive Learning", In: The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol, XXXII, Part 3/1, pp. 367-378.

Contact

Arcot Sowmya (Supervisor)

John Trinder (Supervisor)

Annie Chen (Research Assistant)

Gary Donovan (Research Assistant)