Smart Gauge - Automatic rail survey processing and gauging using deep learning.

点击翻译按钮获取中文摘要

This project, Smart Gauge, will use point cloud data to improve railway surveying and gauging techniques through automation. Our proposed system will automate work towards continuous and instant real time information on rail gauging. The current surveying regime may take place up to every 5 years, GMV plan to bring to market a system which will reduce this internval every 3 to 6 months. Possible future extensions of fidelity through the addition of components such as high precision (GMV proprietary) GNSS technology, and satellite imagery, will assure continuous efficiency after the project ends. We will gain a better perception of up-to-date market technologies and practices used for gauging through our partnership with TfL and Network Rail. We will work together to include expertise assessment and provision of real world data to boost our model performance during Phase 1. This progressive conversation will lead to a strong end-user engagement that will facilitate faster commercialisation and feedback to be the leaders in automatic gauging processing systems in the market. We will meet the technical challenges by utilising technologies at the forefront of Geometric Deep Learning over the last few years, particularly through investment from the self-driving car industry. The two main objectives to be addressed during Phase 1 are: to categorise 10 types of structure and vegetation; and to accurately record cant, curvature and clearance in 5 metre slices of point cloud and write these data to SC0 compliant with the National Gauging Database standard. To fulfil these objectives, our model will perform object identification and segmentation of 3D point clouds. To summarise, the proposed integrated model will remove the major bottleneck currently faced in industry (and highlighted as a key driver for this competition by Network Rail); so that they benefit from a safer railway for passengers and freight, through asset management; financial efficiency in terms of tedious manual labour expenditure; and access to an intuitive interface that brings new data sources and leads the way in predictive maintenance.