Lane Network Generation from Bird's-Eye View Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating road maps from images require extensive manual effort and cost, and fail to accurately determine the connection relationships between lanes at intersections where vehicles have not passed, leading to incomplete data.
Innovation Solution
An apparatus and method that utilize a processor to detect stop lines and intersection areas from bird's-eye view images, identify entry and exit lanes based on stop line ratios, and generate a lane network to connect roads, employing classifiers and skeletonization techniques to automatically extract connection relationships between lanes at intersections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If vehicles actually travel on roads to obtain video images for generating maps, then the map generation uses real trajectory data, but it requires extremely many man-hours and costs
Solution Approach 1:
The system performs preliminary actions by using pre-collected trajectory data from vehicles that have already passed through intersections. Instead of requiring new vehicles to travel specifically for map generation, the system utilizes existing trajectory information that has been collected during normal vehicle operations, thereby avoiding the time-consuming process of deploying vehicles solely for data collection purposes.
Solution Approach 2:
The system creates a virtual representation of the road network by generating a bird's-eye view image and detecting lane structures from collected trajectory data. This copying approach allows the system to reconstruct the road geometry and lane connections without physically traveling on the roads, significantly reducing the time and resources required while maintaining measurement accuracy.
2Reliability
If statistical analysis of vehicle trajectories is used to obtain connectable entry and exit points, then real-world usage data is utilized, but the connection relationship cannot be obtained for intersections that no vehicle has actually passed
Solution Approach 1:
The system performs preliminary detection of stop lines and intersection areas from bird's-eye view images before attempting to determine lane connections. By pre-identifying the geometric structure of intersections and the positions of stop lines, the system can infer potential lane connections even for intersections without trajectory data, thereby expanding coverage while maintaining reliability through subsequent validation.
Solution Approach 2:
The system introduces bird's-eye view images and detected geometric features (stop lines, intersection areas, lane structures) as intermediary elements between the available trajectory data and the target lane connection information. These intermediaries allow the system to infer connections at intersections without direct vehicle observations by using the detected road geometry and stop line positions as intermediate representations that bridge the gap between partial trajectory data and complete intersection models.
3Extent of automation
If the system detects entry and exit lanes based on stop line ratios, then automated lane identification is achieved, but complex intersection geometries may be misclassified
Solution Approach 1:
The system applies local quality by using different detection strategies for different parts of the intersection based on the detected stop line positions and road geometry. Instead of applying a uniform classification rule throughout, the system adjusts the lane detection approach according to the specific local characteristics of each road segment and intersection area, thereby improving classification accuracy while maintaining automation.
Solution Approach 2:
The system performs preliminary detection of stop lines and road geometries before classifying entry and exit lanes. By first establishing the geometric framework and stop line positions, the system creates a structured basis for lane classification that reduces misclassification errors, allowing automated detection to proceed with higher accuracy even in complex intersection geometries.
Data Source
AI summary
An apparatus for generating a map includes a processor configured to detect a stop line of an intersection from a bird's-eye view image, detect from the bird's-eye view image an intersection area including an intersection and roads connected to the intersection, detect at least either one of an entry lane for entering the intersection and an exit lane for exiting the intersection in a road from which the stop line is extracted, of the roads, based on the ratio of the length of the stop line to the width of the road, and generate a lane network representing a connection relationship between lanes in the intersection so as to connect, for each of the roads, the entry lane of the road to the exit lane of another of the roads to which a vehicle is allowed to proceed from the entry lane of the road.