Neural Network Coordinate Conversion for Autonomous Vehicle Positioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The transition of 2D image domain coordinates to 3D world domain coordinates using homography operations in automatic driving systems often results in significant errors, particularly in areas with varying road altitudes, leading to instability in vehicle positioning and navigation.
Innovation Solution
An image processing method that utilizes neural networks to detect objects within driving images, calculates altitude differences, and converts image domain coordinates into world domain coordinates, incorporating pitch information and vanishing line data to improve accuracy, while also generating live maps for dynamic objects and landmark maps for still objects to guide vehicle control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If homography operation is used to convert 2D image domain coordinates to 3D world domain coordinates, then the conversion process is simple and fast, but the positioning accuracy deteriorates significantly in areas with varying road altitudes
Solution Approach 1:
The patent changes the parameters used in coordinate conversion from simple homography transformation to a more complex model that incorporates altitude information. By introducing altitude as an additional parameter and using multi-layer perceptron neural networks, the system achieves accurate 3D coordinate conversion even in areas with varying road altitudes, resolving the contradiction between simplicity and accuracy.
2Measurement precision
If neural network processing is used to determine world domain coordinates with altitude information, then positioning accuracy is improved, but the processing complexity and computational requirements increase
Solution Approach 1:
The patent performs preliminary actions by pre-processing image data to extract relevant features and pre-calculating altitude differences. By preparing data in advance and using pre-trained neural networks, the system reduces real-time processing complexity while maintaining high positioning accuracy, effectively managing the trade-off between accuracy and processing complexity.
3Reliability
If altitude difference calculation is incorporated into the coordinate conversion process, then correctness in varying altitude environments is improved, but the processing time and computational load increase
Solution Approach 1:
The patent replaces traditional mechanical coordinate conversion methods with neural network-based processing. The neural networks automatically learn and adapt to varying altitude conditions, providing reliable coordinate conversion without requiring complex manual calculations or iterative adjustments, thus maintaining reliability while reducing processing time.
Data Source
AI summary
An image processing method and an image processing device detect an object from a driving image of a vehicle, obtain information on an altitude difference between the vehicle and the object, and input image domain coordinates of the object in the driving image and the information on the altitude difference to a neural network and determine world domain coordinates of the object.