Lane Line Data Processing Using Deep Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current lane line data processing in high-precision maps relies heavily on manual extraction, which is inefficient and labor-intensive, especially as the demand for mass production increases, and is prone to errors due to interference from characters and vehicles in point cloud data.
Innovation Solution
A method using a pre-trained deep neural network model to extract lane line areas and attributes from two-dimensional grayscale images derived from three-dimensional point cloud data, followed by splicing these images to obtain lane line data, incorporating key point sampling and filtering to enhance accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual extraction is used to obtain lane line data, then the accuracy of lane line extraction can be maintained, but the processing efficiency is low and labor costs are high
Solution Approach 1:
The patent replaces the manual mechanical extraction process with an automated deep learning system. A neural network model is trained to automatically identify and extract lane lines from point cloud data, substituting human operators with an intelligent algorithm that can process data continuously without fatigue or interruption, thereby dramatically improving processing efficiency
Solution Approach 2:
The system enables self-service extraction where the deep learning model autonomously performs lane line identification and data extraction without requiring manual intervention. The model processes point cloud data independently, automatically generating lane line information that would traditionally require human analysts to interpret and extract
2Ease of manufacture
If manual extraction is used to obtain lane line data, then the processing can be completed, but the labor costs are high
Solution Approach 1:
The patent replaces human labor with an automated deep learning system that performs lane line extraction. The neural network model processes point cloud data and generates lane line information automatically, eliminating the need for human operators to manually interpret and extract lane lines from complex point cloud data, thereby reducing labor resources required
3Measurement precision
If point cloud data is processed directly, then the complete information is available, but interference from characters and vehicles reduces extraction accuracy
Solution Approach 1:
The patent extracts only the relevant lane line information from the point cloud data while filtering out interfering elements such as characters and vehicles. The deep learning model is trained to specifically identify lane line patterns and separate them from other road elements, effectively extracting the needed information while discarding harmful interference
Solution Approach 2:
The patent applies different processing characteristics to different regions of the point cloud data. The deep learning model learns to identify specific local patterns characteristic of lane lines versus other road elements, applying appropriate extraction techniques to each region based on its local characteristics, thereby improving accuracy by treating different areas with specialized approaches
4Measurement precision
If the entire road image is processed at once, then the complete lane line data can be obtained, but the processing time increases
Solution Approach 1:
The patent divides the road image or point cloud data into multiple segments or regions that can be processed independently and in parallel. The deep learning model processes each segment separately, and the results are then integrated to form the complete lane line data. This segmentation approach maintains data completeness while reducing overall processing time through parallel computation
Data Source
AI summary
The present disclosure describes methods, device, and storage medium for obtaining lane line data of a road. The method includes obtaining and dividing, by a device, road-image data of a road into at least one segment. The device includes a memory storing instructions and a processor in communication with the memory. The method includes processing, by the device, each segment to obtain a two-dimensional grayscale image of each segment; extracting, by the device using a pre-trained deep neural network model, a lane line area and a lane line attribute in each two-dimensional grayscale image, to obtain a lane line area image of each segment; and splicing, by the device, each lane line area image based on corresponding road-image data and the lane line attribute, to obtain lane line data of the road.