Lane Boundary Detection Using Multi-Image Distance Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles and driver assistance systems face challenges in accurately determining lane boundaries and estimating distances due to the impact of vehicle camera orientation on horizon location within captured images.
Innovation Solution
A method and system for lane detection that receives initial lane boundary estimates, including first and single-image based estimates, and generates real-world lane detection estimates by evaluating distances between these estimates associated with the same point in time, while compensating for vehicle movement and camera orientation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If single-image based lane boundary estimates are used, then the system can provide lane detection under various driving conditions, but the accuracy is reduced due to camera orientation impact on horizon location
Solution Approach 1:
The patent combines multiple lane boundary estimate sources (first lane boundary estimates from multiple images and single-image based lane boundary estimates) into a unified lane detection model. This merging allows the system to leverage the adaptability of single-image estimates while compensating for their accuracy limitations through integration with multi-image data and vehicle state information.
Solution Approach 2:
The patent introduces vehicle speed and yaw angle as additional parameters to compensate for camera orientation effects. By incorporating these dynamic parameters, the system adjusts the lane boundary detection to account for horizon location shifts caused by vehicle movement, thereby maintaining accuracy across varying driving conditions.
2Measurement precision
If multiple images are processed to improve lane boundary accuracy, then measurement precision increases, but processing time and computational complexity increase
Solution Approach 1:
The patent performs preliminary processing by evaluating real-world distances between initial lane boundary estimates from multiple images before final integration. This preliminary evaluation filters and prepares data in advance, reducing the computational burden during real-time lane detection and minimizing processing time while maintaining accuracy.
Solution Approach 2:
The patent segments the lane detection process into distinct stages: acquiring multiple images with lane boundary estimates, evaluating real-world distances between these estimates, and integrating them into a final lane detection model. This segmentation allows each stage to be optimized independently, balancing accuracy requirements with processing efficiency.
3Measurement precision
If vehicle speed and yaw angle information are incorporated to compensate for camera orientation, then lane detection accuracy improves, but device complexity increases
Solution Approach 1:
The patent integrates vehicle speed and yaw angle information into the existing lane detection system, allowing these parameters to serve multiple functions: compensating for camera orientation effects, improving distance estimation accuracy, and adapting to various driving conditions. This multi-functional use of vehicle state data improves accuracy without requiring separate dedicated systems.
Data Source
AI summary
A method for lane detection, including A computer implemented method for lane detection, the method includes (a) receiving, at one or more processing circuits of a vehicle, a plurality of initial lane boundary estimates that represent lane boundaries within a road environment, the plurality of initial lane boundary estimates include (i) first lane boundary estimates and (ii) single-image based lane boundary estimates; wherein under one or more predefined conditions the single-image based lane boundary estimates are sent once per multiple images; and (a) generating, by the one or more processing circuits, real-world lane detection estimates based on the initial lane boundary estimates, the generating includes evaluating real-world distances between initial lane boundary estimates of the lane boundaries, the initial lane boundary estimates are associated with a same point of time.