LIDAR Intensity Localization Using Gaussian Mixture Map Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current localization systems for autonomous vehicles face challenges in achieving accurate and efficient localization, particularly in situations where GPS provides insufficient accuracy and existing methods are computationally intensive, limiting the vehicle's range and speed of localization.
Innovation Solution
The method employs LIDAR sensor data to extract a Gaussian Mixture Model (GMM) distribution of intensity values using expectation-maximization, calculating a log-likelihood for sub-regions of a prior map to determine vehicle location, which is computed in real-time, reducing computational overhead and storage constraints, and enabling centimeter-level accuracy even in adverse conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional localization methods are used, then localization accuracy may be sufficient, but computational overhead is high and vehicle range is limited
Solution Approach 1:
The prior map is divided into multiple sub-regions, and the system only processes the relevant sub-region containing the vehicle's hypothesis position. This segmentation reduces the amount of data that needs to be processed from the entire map to just a localized area, thereby reducing computational overhead while maintaining localization accuracy.
Solution Approach 2:
The patent extracts only the necessary features (intensity values from LIDAR sensor data) from the sensor data and prior map, discarding unnecessary information. By extracting and comparing only the relevant intensity values in the sub-region, the system achieves accurate localization with reduced computational requirements.
2Reliability
If traditional localization methods are used, then localization can be performed, but storage constraints are significant
Solution Approach 1:
The system extracts only the essential intensity values from LIDAR sensor data and the prior map sub-region, rather than storing and processing complete point cloud data or full map information. This extraction approach maintains localization reliability by preserving the key features needed for comparison while dramatically reducing storage requirements.
Solution Approach 2:
The patent uses lightweight intensity value representations instead of heavy point cloud data structures. These simplified intensity values act as disposable, computationally efficient proxies for the full sensor data, enabling reliable localization with minimal storage constraints.
3Measurement precision
If GPS is used for localization, then coverage area is large, but accuracy is insufficient for autonomous driving
Solution Approach 1:
The patent combines LIDAR sensor data from the vehicle with pre-stored prior map data to create a hybrid localization system. This merging of real-time sensor observations with historical map information enables the system to achieve both high accuracy (like GPS-denied methods) and environmental coverage (by using the extensive prior map data), resolving the contradiction between precision and versatility.
4Measurement precision
If computationally intensive localization methods are used, then accuracy can be maintained, but vehicle speed and operational efficiency decrease
Solution Approach 1:
By segmenting the prior map into sub-regions and only processing the relevant sub-region around the vehicle's hypothesis position, the system maintains localization accuracy through detailed comparison while dramatically reducing computation time. This enables fast localization that keeps pace with vehicle speed requirements.
Solution Approach 2:
The system performs partial action by comparing only the essential intensity values in the relevant sub-region rather than processing all possible data. This partial processing approach achieves sufficient accuracy for autonomous driving without the excessive computational burden of complete data analysis, thereby improving localization speed.
Data Source
AI summary
Systems, methods, and devices for determining a location of a vehicle or other device are disclosed. A method includes receiving sensor data from a sensor and determining a prior map comprising LIDAR intensity values. The method includes extracting a sub-region of the prior map around a hypothesis position of the sensor. The method includes extracting a Gaussian Mixture Model (GMM) distribution of intensity values for a region of the sensor data by expectation-maximization and calculating a log-likelihood for the sub-region of the prior map based on the GMM distribution of intensity values for the sensor data.