Lidar Point-Cloud Clustering for Reliable Vehicle Ego-Motion Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining vehicle ego motion using lidar sensors are susceptible to errors due to dynamic and interfering objects, particularly in urban environments, where guide rails and moving vehicles complicate accurate estimation.
Innovation Solution
A method that filters out dynamic and interfering objects by clustering and selecting points based on shape and spatial location within point clouds, using techniques like DBSCAN and principal component analysis, followed by an iterative closest point algorithm to determine displacement and estimate ego motion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the iterative closest point method is used with all points in the point cloud, then the ego motion estimation can be performed, but dynamic and interfering objects cause errors and reduce reliability
Solution Approach 1:
The point cloud is segmented into multiple clusters based on spatial proximity, where each cluster represents a potential object or group of points. This segmentation allows the system to process and evaluate individual clusters separately, identifying and excluding dynamic or interfering objects while maintaining the overall structure needed for ego motion estimation.
Solution Approach 2:
Different quality criteria are applied to different clusters based on their local characteristics. Clusters are evaluated individually using shape descriptors and point density metrics, allowing the system to selectively use high-quality clusters for ego motion estimation while filtering out problematic ones, thereby improving reliability without uniformly increasing complexity.
2Reliability
If simple point density filtering is used to remove outliers, then processing is simple, but interfering objects with high point density like guide rails are not recognized and cause errors
Solution Approach 1:
Shape descriptors are calculated for each cluster before the ego motion estimation process. These descriptors capture the geometric characteristics of clusters in advance, allowing the system to pre-identify and filter out clusters from dynamic or interfering objects before they can negatively impact the estimation, thereby improving accuracy without adding complex real-time detection mechanisms.
Solution Approach 2:
The system uses shape descriptors as additional parameters beyond simple point density. By incorporating geometric parameters that describe the form and structure of clusters, the system can distinguish between static objects suitable for ego motion estimation and dynamic or interfering objects, even when the latter have high point density like guide rails.
3Productivity
If clusters with low point density are filtered out using the outlier method, then processing is simplified, but valid objects with sparse points are incorrectly removed
Solution Approach 1:
Each cluster is evaluated using multiple quality criteria including shape descriptors and point density, rather than relying solely on point density. This localized quality assessment allows the system to retain clusters from valid objects even when they have sparse points, while still filtering out true outliers, thereby maintaining measurement precision without sacrificing processing efficiency.
Solution Approach 2:
The quality assessment of clusters combines multiple parameters (shape descriptors, point density, spatial distribution) into a composite evaluation metric. This composite approach allows the system to make more informed decisions about which clusters to retain, preventing the incorrect removal of valid objects with sparse points while maintaining efficient processing through systematic filtering.
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
Enhances the reliability and accuracy of ego motion estimation by filtering out irrelevant objects, allowing for precise determination using low-cost lidar sensors even in complex environments.
Implementation Method 1
light signals can be emitted by the lidar sensor and the light signals reflected in an environment of the vehicle and in particular on objects can be received again
Implementation Method 2
On the basis of the time-of-flight between the emission of the respective light signal and the reception of a light signal reflected by the object, the distance between the lidar sensor and the object can then be determined
Data Source
AI summary
A method estimates an ego motion of a vehicle. The method includes the steps of: determining a first point cloud having a plurality of first points on the basis of sensor data which are provided by a lidar sensor of the vehicle at a first point in time; determining a second point cloud having a plurality of second points on the basis of sensor data which are provided by the lidar sensor at a second point in time following the first point in time; associating the first points to clusters and the second points to clusters; selecting first points from the first point cloud and selecting second points from the second point cloud; determining a displacement between respective corresponding selected first points; and selected second points and estimating the ego motion of the vehicle on the basis of the displacement, wherein the selection of the first points and the selection of the second points are carried out according to a shape of the cluster.