3D Rail Model Adjustment via Camera Activation Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting rail tracks in automated and driverless train systems are either independent of calibration but prone to errors or require additional sensors, which are error-prone and require sophisticated calibration, and do not accurately determine the location or dimensions of the tracks.
Innovation Solution
A method for determining an adjusted parameterized 3D rail model using a single image recording unit, which generates an activation map from image data to adapt a generalized 3D rail geometry model to the actual rail track course, allowing for precise detection of rail poses and handling sparse or noisy data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If image-based approaches with edge filtering or machine learning are used to detect the track, then the system remains independent of calibration and avoids additional system-related errors, but the methods do not determine the position or precise dimensions of the tracks accurately
Solution Approach 1:
The patent introduces an auxiliary camera as an intermediary sensor to capture images of the track geometry. This auxiliary camera works in conjunction with the main object recognition camera to provide additional geometric information. The auxiliary camera is calibrated relative to the main camera, creating a coordinated sensor system that maintains calibration independence while improving measurement precision through multi-view geometry.
Solution Approach 2:
The patent transitions from two-dimensional image-based track detection to three-dimensional track geometry determination by incorporating depth information from the auxiliary camera. This dimensional enhancement allows the system to accurately determine track position and dimensions without relying on complex calibration of a single sensor, as the stereo-like configuration provides inherent geometric constraints.
2Measurement precision
If additional sensors such as auxiliary cameras or Lidar are added to extract the track, then more accurate track detection is achieved, but the system becomes more complex and requires sophisticated calibration
Solution Approach 1:
The auxiliary camera is designed to serve multiple functions: it captures images for track geometry extraction, provides depth information for 3D reconstruction, and enables accurate position and dimension determination. This multi-functionality justifies the addition of the sensor while maximizing its utility across different detection tasks.
Solution Approach 2:
The patent changes the calibration parameters from complex multi-sensor calibration to a simplified relative calibration between two cameras. By focusing on determining the transformation between the main camera and auxiliary camera coordinate systems, the system achieves accurate track detection with reduced calibration complexity compared to traditional multi-sensor setups.
3Measurement precision
If INS-based methods with precise maps are used, then direct inferences about track alignment are possible, but map creation is expensive and laborious and the map can quickly become outdated
Solution Approach 1:
The system performs self-calibration and self-updating by continuously capturing images of the track geometry and automatically adjusting its internal model. Instead of relying on pre-created maps that require external maintenance, the system serves itself by using real-time image data to update its understanding of track alignment, eliminating the need for expensive and laborious map creation and updates.
Solution Approach 2:
The patent performs preliminary calibration between the main camera and auxiliary camera once, establishing the transformation relationship before actual track detection begins. This preliminary action simplifies subsequent operations, as the system only needs to process image data without requiring complex real-time calibration or external map data, thereby avoiding the ongoing maintenance burden of traditional INS-based systems.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
A method for determining an adjusted parameterized 3D rail model (M') is described. The method includes acquiring image data (BD) of an environment (UB) of a rail vehicle (1), which comprises a track (ST), by an image acquisition unit (BA) arranged on the rail vehicle (1). Based on the acquired image data (BD), an activation map (AK) is determined. Finally, the adjusted parameterized 3D rail model (M'), which represents the pose (P) of the track (ST) relative to the image acquisition unit (BA), is determined by comparing a parameterized 3D rail model (M) with the determined activation map (AK). An adjustment device (40) is also described. Furthermore, a rail vehicle (1) is described.