LiDAR Landmark Perception With DNN Curve Fitting for AV Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional perception and localization methods for autonomous vehicles are limited in functionality, accuracy, and precision when using LiDAR data only or detecting certain landmarks like lane lines, particularly in curved road environments, due to sparse LiDAR points and inefficiencies in techniques like the Hough transform.
Innovation Solution
The use of deep neural networks (DNNs) to classify LiDAR points, regress landmark shapes, and generate binary masks, which are then clustered and fitted to represent detected landmarks for localization and navigation, enabling more accurate detection of lane lines and other objects using LiDAR data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If per-point classification of LiDAR points is used to detect landmarks, then localization capability is enabled, but localization precision is insufficient due to sparse LiDAR points
Solution Approach 1:
The patent combines multiple LiDAR point classifications within a local neighborhood to form a collective landmark representation. By merging adjacent classified points that belong to the same landmark instance, the system creates denser and more reliable landmark detections, overcoming the sparsity issue of individual LiDAR points while maintaining localization capability.
2Ease of manufacture
If Hough transform is used to find lines in classifications, then straight line detection is achieved, but efficiency decreases and correctness becomes difficult to ensure when roads have curves
Solution Approach 1:
The patent changes the detection parameters by using fitted curves with adjustable degree parameters instead of fixed straight-line parameters. This allows the detection algorithm to adapt to both straight roads and curved roads by modifying the curve degree, maintaining high efficiency while improving correctness across different road geometries.
Solution Approach 2:
The system transitions from static straight-line detection to dynamic curve fitting that can adapt to varying road geometries. The curve fitting approach dynamically adjusts to the actual landmark shape in the scene, whether straight or curved, making the detection process efficient and accurate for diverse road conditions.
3Adaptability or versatility
If Hough transform is generalized to handle curved roads, then curve detection capability is improved, but computational efficiency decreases
Solution Approach 1:
The patent segments the curve fitting process into manageable steps: first identifying candidate regions through point classification, then fitting curves only to relevant segments. This segmentation approach enables curve detection capability while maintaining computational efficiency by avoiding exhaustive processing of entire point clouds.
Data Source
AI summary
In various examples, one or more DNNs may be used to detect landmarks (e.g., lane lines) and regress a representation of their shape. A DNN may be used to jointly generate classifications of measured 3D points using one output head (e.g., a classification head) and regress a representation of one or more fitted shapes (e.g., polylines, circles) using a second output head (e.g., a regression head). In some embodiments, multiple DNNs (e.g., a chain of multiple DNNs or multiple stages of a DNN) are used to sequentially generate classifications of measured 3D points and a regressed representation of the shape of one or more detected landmarks. As such, classified landmarks and corresponding fitted shapes may be decoded and used for localization, navigation, and/or other uses.