3D Lane Line Generation from Point Clouds and Vehicle Direction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating three-dimensional lane lines in automatic drive vehicles suffer from inaccuracies due to mistaken detection of two-dimensional lane lines, leading to broken or disconnected lines, which reduces the accuracy of the generated three-dimensional lane lines.
Innovation Solution
A method that determines lane line key points from both two-dimensional lane line data and three-dimensional point cloud data, utilizing attribute information and driving direction to connect these points randomly and iteratively refine the lane lines based on cost values to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If two-dimensional lane lines are projected into three-dimensional space to generate three-dimensional lane lines, then the generation process is simple, but the accuracy of the generated three-dimensional lane lines deteriorates due to mistaken detection and broken connections
Solution Approach 1:
The patent segments the lane line generation process into multiple stages: first extracting two-dimensional lane line data from images, then projecting it to three-dimensional space, and finally refining it through iterative optimization. This segmentation allows each stage to be optimized independently, improving overall accuracy while maintaining computational efficiency.
Solution Approach 2:
The patent implements a feedback mechanism through iterative optimization where the generated three-dimensional lane lines are continuously refined based on consistency checks with the original two-dimensional detections and spatial relationships. This feedback loop corrects mistakes and breaks in the lane lines, significantly improving accuracy.
2Speed
If two-dimensional images of adjacent frames are projected onto three-dimensional ground to obtain lane lines, then the processing speed is fast, but the lane lines become broken or disconnected reducing accuracy
Solution Approach 1:
The patent performs preliminary actions by pre-processing the two-dimensional lane line detections to identify and mark potential breaks or inconsistencies before the projection step. This allows the subsequent projection and refinement processes to focus only on correcting these pre-identified issues, maintaining speed while improving reliability.
Solution Approach 2:
The iterative optimization process provides feedback that continuously checks the continuity and consistency of the projected lane lines, allowing corrections to be made in real-time during the projection process rather than requiring separate post-processing steps.
3Measurement precision
If lane line key points are connected randomly multiple times to refine the lane line, then the accuracy of three-dimensional lane line improves, but the computational complexity increases
Solution Approach 1:
The patent applies local quality by focusing the iterative optimization on specific segments of the lane line where breaks or inconsistencies are detected, rather than uniformly processing the entire lane line. This localized approach reduces computational complexity while maintaining high accuracy in the critical areas.
Solution Approach 2:
The patent changes parameters such as the number of iteration steps, the threshold for accepting random connections, and the weighting of different cost functions to balance accuracy and computational complexity. This allows optimization of the trade-off based on specific application requirements.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
The disclosure relates to a method for generating a lane line, a vehicle, a storage medium and a chip. The method includes: acquiring three-dimensional point cloud data and two-dimensional lane line data of a target lane, the two-dimensional lane line data being lane line data collected in response to determining that the vehicle is driving in the target lane, and the three-dimensional point cloud data being point cloud data of the surrounding environment of the vehicle collected in response to determining that the vehicle is driving in the target lane; determining a plurality of lane line key points according to the two-dimensional lane line data and the three-dimensional point cloud data; and obtaining a target three-dimensional lane line by connecting the plurality of lane line key points randomly for many times according to attribute information of the plurality of lane line key points and a driving direction of the vehicle.