Road extraction in RGB images acquired by Low Altitude Remote Sensing from an Unmanned Aerial Vehicle: A Neural Network Based Approach

摘要

The high growth rate of urban population has led to an increase in the demand for better urban planning and monitoring which mainly includes road network development. Manual monitoring of road development is time consuming and inefficient. In this paper, we propose a method for automatic extraction of roads in vision spectrum (RGB) images acquired by remote sensing from a UAV also known as Low Altitude Remote Sensing (LARS) or Near Earth Remoste Sensing. Extreme Learning Machine (ELM), a neural networks based classier is used for spectral classication. Spectral classication is further improved by applying spatial techniques. The spatial techniques include a combination of Shape Index(SI), Density Index(DI) and mathematical morphological close operations. Seven images of diverse road stretches are analyzed to verify the robustness of the proposed method. The classification results are analysed using confusion matrix. The performance parameters derived from confusion matrix are analyzed for a range of hidden neurons of the ELM model and an optimum number of hidden neurons are chosen. Successful road extraction demonstrates the potential of using UAV imagery for monitoring road development.