Visual and LIDAR Data Processing and Fusion as an Element of Real Time Big Data Analysis for Rail Vehicle Driver Support Systems

Abstract

This chapter reviews the challenges, processing and analysis techniques about visual and LIDAR generated information and their potential use in big data analysis for monitoring the railway at onboard driver support systems. It surveys both sensors' advantages, limitations, and innovative approaches for overcoming the challenges they face. Special focus is given to monocular vision due to its dominant use in the field. A novel contribution is provided for rail extraction by utilizing a new hybrid approach. The results of this approach are used to demonstrate the shortcomings of similar strategies. To overcome these disadvantages, dynamic modeling of the tracks is considered. This stage is designed by statistically quantifying the assumptions about the track curvatures presumed in current railway extraction techniques. By fitting polynomials to hundreds of manually delineated video frames, the variations of polynomial coefficients are analyzed. Future trends for processing and analysis of additional sensors are also discussed.