3D Lane Detection Using Anchor Grids for Curved Road Markings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing lane detection technologies face challenges in accurately detecting curved lane markings, especially at intersections and areas where lanes cross, leading to unreliable results due to the need for heuristic combination of relatively straight lane segments.
Innovation Solution
An end-to-end deep learning approach is used to generate 3D points of lane markings in real-time by training a neural network with known 3D points relative to anchor points, which are centers of concentric circles, allowing for accurate representation and detection of lane markings regardless of curvature.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional lane detection methods are used for curved lane markings, then detection complexity increases, but detection accuracy deteriorates
Solution Approach 1:
The patent transforms the lane detection problem from 2D image space to 3D space by introducing depth information through a grid of anchor points with varying distances from the vehicle. This dimensional transformation allows the neural network to directly predict 3D lane parameters (position, orientation, curvature) without complex heuristic processing of curved segments, thereby improving accuracy while maintaining system simplicity
Solution Approach 2:
The patent replaces traditional mechanical/heuristic lane detection algorithms with a data-driven neural network approach. Instead of using geometric rules and heuristic combinations to detect curved lanes, the system uses a trained neural network that learns optimal detection patterns from training data, simplifying the detection mechanism while improving accuracy for curved and intersecting lanes
2Reliability
If heuristic combination of straight lane segments is used, then detection reliability deteriorates, but computational resources are consumed
Solution Approach 1:
The patent performs preliminary action by pre-defining a grid of anchor points covering the entire road area before detection. These anchor points with known 3D coordinates serve as reference markers that enable direct prediction of lane parameters without needing to combine multiple straight segment detections. The neural network uses these pre-positioned anchors to reliably detect curved lanes in a single pass, improving reliability while reducing computational energy compared to iterative heuristic methods
3Reliability
If 3D lane detection is implemented, then navigation safety improves, but system complexity increases
Solution Approach 1:
The patent segments the road space into a grid of discrete anchor points, each with predefined 3D coordinates. This segmentation transforms the continuous 3D space into manageable discrete units that the neural network can efficiently process. By detecting which anchor points lie on lane markings and their spatial relationships, the system achieves reliable 3D lane detection without requiring complex continuous 3D reconstruction algorithms, thus improving navigation safety while controlling system complexity
Data Source
AI summary
Systems and methods to perform image-based three-dimensional (3D) lane detection involve obtaining known 3D points of one or more lane markings in an image including the one or more lane markings. The method includes overlaying a grid of anchor points on the image. Each of the anchor points is a center of i concentric circles. The method also includes generating an i-length vector and setting an indicator value for each of the anchor points based on the known 3D points as part of a training process of a neural network, and using the neural network to obtain 3D points of one or more lane markings in a second image.