Accuracy Assessment of Automatic Road Features Extraction from Unmanned Autonomous Vehicle (UAV) Imagery

Abstract

Nowadays, the road network documentation is required for many applications, it is essential for the development of economy and its growth brings benefits in people’s life. Traditionally, the road network extraction is done manually, however, it is costly, and time consuming to update and utilize the spatial information. Thus, in order to utilize this issue, this study aims to evaluate the capabilities of automatic road extraction from orthophoto UAV images using Trainable Weka Segmentation (TWS), Level Set (LS) and Seeded Region Growing (SRG) methods. The study area was carried out at UiTM Perlis Branch area. In this study, The UAV image was processed by using Agisoft PhotoScan software to produce orthophoto image, then the road network in the orthophoto was segmented and extracted by using ImageJ Fiji. Several ground controls were also established at the surrounding of study area. For validation purposes, the automatic extracted road network was compared against manual extracted road network. Based on the findings, it was found that SRG method is slightly better in extracting road features compared to LS method in term of completeness, correctness, and quality for automated extraction. It is hope, this study can be used to help reducing the cost and time consumed in extracting features, especially road network, by using automatic extraction instead of manual extraction.