LiDAR Lane Recognition Using Point-Cloud Learning in Variable Light
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Lane recognition technologies using cameras face challenges in environments with extreme illuminance changes, such as high or low surrounding light, and backlight conditions, leading to reduced recognition accuracy.
Innovation Solution
A LiDAR-based lane recognition system comprising a pre-processor that extracts lane point candidates from LiDAR data, a deep learning device that uses multilayer perceptron-based learning to classify points, and a post-processor for curve fitting, which samples and processes point cloud information to enhance lane detection accuracy across varying illuminance conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If camera-based lane recognition is used, then the system is simple and cost-effective, but the recognition accuracy deteriorates in extreme illuminance conditions
Solution Approach 1:
The patent replaces the camera-based optical detection system with a LiDAR-based laser ranging system. LiDAR uses laser beams to measure distances and generate point cloud data, which is inherently immune to illuminance changes. This substitution of detection mechanism fundamentally eliminates the sensitivity to lighting conditions while maintaining lane recognition capability.
Solution Approach 2:
The patent changes the detection parameter from optical intensity (camera) to time-of-flight distance measurement (LiDAR). By measuring the time for laser pulses to travel to and from targets, the system obtains depth information that is independent of ambient light conditions, thereby resolving the contradiction between accuracy and illuminance sensitivity.
2Measurement precision
If LiDAR-based lane recognition is implemented, then the recognition accuracy in varying illuminance is improved, but the device complexity increases
Solution Approach 1:
The patent segments the lane recognition task into distinct processing stages: point cloud data acquisition from LiDAR, point cloud processing and filtering, lane line detection from processed points, and curve fitting for final lane representation. This segmentation allows each module to be optimized independently and simplifies the overall system architecture despite using LiDAR technology.
Solution Approach 2:
The patent extracts and processes only the relevant point cloud data corresponding to the road surface and lane markings from the complete LiDAR point cloud. By filtering and selecting specific points that contribute to lane detection while discarding irrelevant data, the system reduces computational complexity while maintaining high recognition accuracy.
3Measurement precision
If point cloud data is processed without sampling, then all data is utilized for maximum accuracy, but the processing time and computational load increase
Solution Approach 1:
The patent applies partial action by selectively processing only a subset of point cloud data that is most relevant for lane detection. Through sampling and filtering operations, the system processes sufficient data to maintain high accuracy while avoiding the computational burden of processing every single point, thus optimizing the trade-off between accuracy and processing time.
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 system effectively maintains high lane recognition accuracy even in environments with sudden changes in illuminance, outperforming camera-based systems by providing precise lane information and reducing errors associated with extreme lighting conditions.
Implementation Method 1
a pre-processor that detects a lane point candidate group based on information detected from the LiDAR
Data Source
AI summary
A lane recognition device based on a Light Detection and Ranging (LiDAR) for a vehicle and a method thereof may include a pre-processor that detects a lane point candidate group based on information detected from the LiDAR, a deep learning device that generates lane recognition information by learning the lane point candidate group, and a post-processor that converts the lane recognition information into lane information by use of a lane curve fitting algorithm.