Ego-Vehicle Georeferenced Mapping with Odometry-GNSS Histogram Voting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for creating georeferenced maps for autonomous vehicles face issues with GNSS discontinuities and accuracy, leading to ghost images and high memory/processing demands, especially in environments with limited or poor GNSS availability.
Innovation Solution
An offline method using dead reckoning for odometry to create an initial map, followed by post-processing with histogram voting to reduce GNSS noise and determine transformation parameters, reducing the need for large memory and processing power.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If filtering-based methods (Kalman-filter) are used to create georeferenced maps, then the map can be created online with continuous updates, but the map contains ghost images of landmarks due to discontinuities in GNSS pose
Solution Approach 1:
The patent segments the vehicle trajectory into multiple overlapping segments, each processed independently to determine local transformation parameters. This segmentation allows the system to handle GNSS discontinuities locally without propagating errors across the entire map, thereby eliminating ghost images while maintaining efficient processing.
Solution Approach 2:
The patent performs preliminary actions by recording the full vehicle trajectory and GNSS data during the mapping phase, then processes this recorded data offline. This preliminary data collection enables post-processing with histogram voting to determine accurate transformation parameters without the constraints of real-time processing, achieving both high accuracy and computational efficiency.
2Manufacturing precision
If SLAM methods are used to create georeferenced maps, then the maps are accurate and free of ghost images, but the memory and processing power consumption is significantly larger
Solution Approach 1:
The patent extracts only the essential transformation parameters (translation and rotation) needed for georeferencing, separating these from the full SLAM state estimation process. By focusing only on the necessary geometric transformations and using histogram voting to determine them, the system achieves SLAM-level accuracy without the computational burden of full SLAM algorithms.
Solution Approach 2:
The patent uses lightweight, computationally inexpensive methods (odometry-based trajectory recording and histogram voting) instead of heavy SLAM algorithms. The system records trajectory data during operation and processes it offline with simple statistical methods, achieving accurate georeferencing without requiring the sustained high computational resources that full SLAM methods demand.
3Duration of action of stationary object
If online filtering methods are used, then the map can be updated continuously, but the GNSS noise causes discontinuities and ghost images in the map
Solution Approach 1:
The patent creates a copy of the vehicle trajectory from odometry data, which is then used as the basis for map creation. This odometry-based trajectory copy is free from GNSS noise and discontinuities, allowing the system to generate consistent, reliable maps without the ghost images that plague filtering-based methods that directly use noisy GNSS pose estimates.
Data Source
Figure 1~2
AI summary
The invention relates to a method for generating a georeferenced map for an ego vehicle (1) comprising the following steps: - Creating (S1) a first map based on sensor data from environmental sensors located in the ego vehicle (1); - Recording (S2) a vehicle trajectory (T1, T2) from odometry and from GNSS in a memory; - Generating (S3) segments (s1, s2, ..., sk) of the vehicle trajectories; - Registering (S4) the generated segments (s1, s2, ..., sk) by determining transformation parameters when matching the segments (s1, s2, ..., sk) of the odometry trajectory (T1) to the segments (s1, s2, ... , sk) of the GNSS trajectory (T2); - Creating (S5) a histogram of the transformation parameters using the histogram voting technique to select the transformation parameters with the highest number of votes, and calculating the number of peaks, their widths and their distances in order to estimate the mapping quality; - Generating (S6) a georeferenced map using the selected transformation parameters.