LiDAR Foreground Extraction Using Surface Fitting on Sloped Terrain
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing LiDAR-based foreground extraction methods for autonomous vehicles are slow and prone to false negatives, particularly in scenarios with sloped road obstacles, as they fail to adapt effectively to terrain changes and often misclassify roadside curbs and foreground objects.
Innovation Solution
The use of a spatial Kalman filter for surface fitting, which transforms LiDAR point cloud data into a 2D polar grid, computes measured and predicted heights and slopes, and employs statistical models to differentiate between foreground and ground points based on innovation covariance, thereby improving the accuracy of foreground extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional 2D Bayesian linear regression (RLS) is used for foreground extraction, then the method is computationally simple, but it is slow and generates false negatives in sloped terrain scenarios
Solution Approach 1:
The patent applies a dynamic terrain model that adapts to changing terrain conditions in real-time, allowing the system to adjust to sloped roads and varying ground surfaces dynamically, thereby improving detection accuracy without sacrificing processing speed
Solution Approach 2:
The system changes key parameters including transforming to a moving reference frame attached to the vehicle, using elevation differences relative to the vehicle, and adapting statistical models based on terrain characteristics, which enables accurate foreground extraction across diverse terrain types
2Adaptability or versatility
If traditional RLS methods are used, then the algorithm is easy to implement, but it fails to adapt to terrain changes and misclassifies roadside curbs and foreground objects
Solution Approach 1:
The patent implements a dynamic reference frame that moves with the vehicle and updates terrain models in real-time, enabling the system to adapt to changing terrain conditions while maintaining a structured algorithmic approach
Solution Approach 2:
The system introduces an intermediate terrain model that acts as a mediator between raw LiDAR data and foreground classification, using statistical models to represent terrain characteristics and improve classification accuracy
3Reliability
If foreground extraction accuracy is improved by using more sophisticated methods, then false negatives are reduced, but processing time increases
Solution Approach 1:
The patent segments the point cloud processing into distinct stages: terrain model estimation, foreground candidate identification, and classification, allowing efficient processing while maintaining high accuracy through focused computational efforts at each stage
Solution Approach 2:
The system performs preliminary terrain model estimation and reference frame transformation before foreground extraction, preparing the data structure in advance to enable faster and more accurate classification in the subsequent processing stage
Data Source
AI summary
Techniques are provided for foreground extraction from a point cloud (e.g., a LiDAR point cloud) using surface fitting. In an embodiment, one or more processors of a vehicle can receive point cloud data from one or more vehicle sensors. The one or more processors can identify points in the point cloud data as foreground points or ground points using a spatial Kalman filter to capture changes in the terrain. A route or trajectory in a driving area for the vehicle can be generated using the identified foreground points and ground points. A vehicle controller can control the vehicle while the vehicle is traveling on the route or trajectory in the driving area.