A Comparison Study of Neural Network-Based Semantic Segmentation for Off-Road Traversability
Title: A Comparison Study of Neural Network-Based Semantic Segmentation for Off-Road Traversability
Authors: Semih Beycimen (Cranfield University); Dmitry Ignatyev (Cranfield University); Argyrios Zolotas (Cranfield University);
Year: 2022
Citation: Beycimen, S., Ignatyev, D., Zolotas, A., (2022). A Comparison Study of Neural Network-Based Semantic Segmentation for Off-Road Traversability. UKRAS22 Conference “Robotics for Unconstrained Environments” Proceedings, 72-73. doi: 10.31256/Wt3Yp1E
Abstract:
Abstract—This paper presents work from a PhD study on unmanned ground vehicle advanced traversability. In particular, in this paper a number of learning algorithm have been trained and tested using the YAMAHA dataset (an off-road related dataset). Results were analysed and compared in terms of prediction accuracy and training time. It was noted that while
various models provide appropriate accuracy results, only few provide results that can be classed as optimal when training time is considered.