Rail Obstacle Detection via Depth Data Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current obstacle detection methods for rail vehicles, particularly in rail traffic, face challenges in identifying diverse obstacles effectively due to the need for extensive and costly training databases, limitations in 2D image processing, and difficulties in real-time processing of large data sets from LIDAR sensors, which restrict their applicability and accuracy, especially for smaller objects and longer distances.
Innovation Solution
A method utilizing 3D image data captured from the surrounding area of a rail vehicle, incorporating disparity images and depth information, projects this data onto a 2D plane aligned with the vehicle's perspective, allowing for rail detection and localization, and identifying potential collision obstacles by comparing depth values, enabling generic obstacle detection without prior knowledge of object shapes or annotations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If supervised object detection with machine learning is used to identify obstacles, then detection effectiveness for most objects is improved, but the cost and complexity of training and annotating databases increases significantly
Solution Approach 1:
The patent extracts and utilizes depth information from 3D image data as a separate, independent feature layer. By projecting 3D data onto a 2D plane and comparing depth values, the system separates depth analysis from traditional 2D image processing, eliminating the need for extensive object-specific training databases while maintaining high detection reliability
Solution Approach 2:
The depth comparison method serves multiple functions: it detects obstacles of any shape or size, identifies their spatial position, determines their distance from the rail vehicle, and operates without requiring pre-trained models for different obstacle types. This universal approach replaces multiple specialized training databases with a single generic depth analysis system
2Productivity
If 2D image data processing is used for obstacle identification, then processing speed is improved, but the ability to determine whether an object is blocking the rails deteriorates, especially in bends
Solution Approach 1:
The patent projects 3D image data onto a 2D plane that is oriented according to the rail vehicle's perspective and the track geometry. This dimensional transformation preserves depth information while adapting the data to a 2D processing format, enabling both fast processing and accurate determination of whether objects block the rails even in curved sections
3Measurement precision
If LIDAR sensors are used to capture 3D image data for obstacle detection, then depth information is improved, but the data processing load increases significantly, making real-time processing difficult
Solution Approach 1:
The system extracts only the essential depth information needed for obstacle detection by comparing depth values at corresponding pixels in the projected 2D plane. This selective extraction approach processes only critical depth data rather than analyzing the entire 3D point cloud, enabling real-time processing while maintaining depth information quality
Solution Approach 2:
Instead of processing all 3D LIDAR data comprehensively, the system performs partial processing by comparing depth values only at pixels that correspond to potential obstacle locations in the projected 2D view. This partial action approach provides sufficient depth information for safety-critical detection while dramatically reducing the computational load to enable real-time operation
Data Source
AI summary
A method for detecting obstacles for a rail vehicle includes capturing 3D image data from an area surrounding the rail vehicle and generating 2D image data from the 3D image data. Rails are detected and localized in the 2D image data. Depth data are determined in the 2D image data based on the 3D image data. The 2D image data are divided into linear image segments, each having a rail section with a constant depth. A depth value of a pixel of a linear image segment outside the rails is then compared with the respective depth value of the rails. If the difference between the depth value of the pixel and the depth value of the rails exceeds a threshold value, the pixel is part of an object projecting above the level of the terrain. Depending on its position and/or movement, the object is considered a potential collision obstacle.