LiDAR Range Image Lane Detection for Sparse Point Clouds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current lane detection systems in autonomous vehicles, particularly those using cameras, are vulnerable to illumination changes and struggle with accurate distance estimation, while LiDAR systems face challenges with sparse point clouds and limited lane detection capabilities due to their dependency on local signal processing.
Innovation Solution
A processor-based apparatus and method that generates a range image from LiDAR point clouds, using a trained lane detection model to extract features, detect lane areas, and acquire data including confidence scores, endpoint values, and polynomial coefficients, enabling accurate lane detection and display on the point cloud.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If camera-based lane detection is used, then lane detection can be performed using image processing algorithms, but the system becomes vulnerable to illumination changes and cannot accurately estimate distance
Solution Approach 1:
The patent combines camera-based lane detection with LiDAR-based distance estimation into a unified system. The camera provides lane marking detection while LiDAR provides accurate depth information, and the two data streams are fused to achieve both illumination robustness and distance accuracy simultaneously
2Reliability
If LiDAR is used for lane detection, then the system is not affected by illumination and can detect lanes in various lighting conditions, but the point cloud becomes sparse and local signal processing methods are limited
Solution Approach 1:
The patent introduces range images as an intermediary representation that transforms sparse LiDAR point cloud data into a dense 2D image format. This range image serves as a mediator that preserves the illumination robustness of LiDAR while enabling the application of sophisticated deep learning algorithms that require dense input data
3Adaptability or versatility
If human-coded heuristic algorithms are applied to LiDAR point clouds, then lane detection can be performed, but it becomes difficult to code for various and all road conditions
Solution Approach 1:
The patent replaces hand-crafted heuristic algorithms with deep learning-based range image processing. Instead of manually coding rules for different road conditions, the system uses neural networks that automatically learn appropriate detection strategies from training data, achieving high adaptability without increasing programming complexity
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The solution enables robust lane detection in various illumination conditions and diverse road scenarios, providing reliable lane information without the need for heuristic algorithms, and improves the accuracy of lane detection by leveraging deep learning algorithms and multi-task loss functions.
Implementation Method 1
The LiDAR detects a distance between a vehicle and a target by emitting a pulse of light to a target point in front of the vehicle and measuring a time for the light pulse to return
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
There is provided methods and apparatuses for detecting a lane using a light detection and ranging (LiDAR), the apparatus including a processor configured to generate a range image based on a LiDAR point cloud acquired from the LiDAR, and acquire at least one lane data by detecting at least one lane area present in the range image using a trained lane detection model.