Cascade Classifier for Road Boundary Detection in 3D Point Clouds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing LiDAR technologies face challenges in classifying road boundaries due to increased complexity and computational load when adding a road boundary classification function, as they are designed for moving objects like vehicles and pedestrians, not static objects like road boundaries.
Innovation Solution
A method using a cascade classifier that combines rule-based and learning-based classification steps, with a rule-based classifier determining candidate road boundaries through box and point parameters, and a learning-based classifier applying a machine-learning scheme using parted covariance features to accurately classify road boundaries, reducing computational load and complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a machine-learning-based classifier is enhanced to include road boundary classification function, then classification capability is improved, but problem complexity increases making learning difficult
Solution Approach 1:
The patent divides the classification problem into two separate stages: a rule-based classification step that filters obvious road boundary candidates using simple geometric rules, and a machine-learning-based classification step that processes only the filtered candidates. This segmentation reduces the complexity of the machine learning problem while maintaining comprehensive classification capability.
2Adaptability or versatility
If conventional classifier functions are extended to include road boundary classification, then classification versatility is improved, but computational load increases
Solution Approach 1:
The patent performs preliminary filtering using rule-based classification before applying the computationally intensive machine learning classifier. By pre-processing the data to identify and filter candidate road boundaries using simple geometric rules, the system reduces the number of points that require complex machine learning analysis, thereby reducing overall computational load.
3Measurement precision
If feature set is expanded to accommodate road boundary classification, then classification accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent applies different feature extraction and classification strategies to different types of objects. For road boundary candidates, it uses specific geometric features and rules tailored to linear structures, while other objects are processed using conventional features. This localized approach improves accuracy for road boundaries without unnecessarily increasing complexity for all classification tasks.
Data Source
AI summary
An embodiment device for detecting a boundary of a road in a 3D point cloud using a cascade classifier includes a rule-based classifier configured to determine whether a received LiDAR cluster has a likelihood of becoming a candidate for the boundary of the road using a box parameter surrounding a point cloud constituting the LiDAR cluster and a point parameter, and a learning-based classifier configured to apply a machine-learning scheme to the LiDAR cluster selected as the candidate for the boundary of the road by the rule-based classifier to determine the LiDAR cluster to be the boundary of the road or an object other than the boundary of the road.