Semantic Point Cloud Mapping for Voice-Guided Vehicle Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional point cloud maps generated by computer vision systems for autonomous vehicles lack semantic information, leading to inaccurate vehicle localization and limited understanding of the environment, especially in indoor areas like covered parking lots.
Innovation Solution
A method and system that integrates human-provided semantic data through voice inputs, processed using natural language processing and an ontology, to generate a semantic point cloud map by associating semantic labels with LiDAR-based point cloud data, enhancing the vehicle's spatial awareness and localization accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If conventional point cloud map generation methods are used, then the map covers a large area with extended boundaries, but the map lacks semantic information and context
Solution Approach 1:
The patent merges LiDAR point cloud data with soft data from human-provided voice inputs by transforming the unstructured soft data into structured semantic data using NLP and ontology. This combination enriches the point cloud map with semantic labels while maintaining its spatial coverage, resolving the contradiction between map area and semantic information content.
Solution Approach 2:
The patent introduces an intermediary processing system that includes NLP modules and ontology structures. This intermediary transforms human voice inputs into structured semantic data that can be integrated with LiDAR point cloud data, enabling the addition of semantic information without compromising the existing map coverage capabilities.
2Loss of information
If soft data from human voice inputs is integrated into the point cloud map, then semantic information and context are improved, but the system complexity increases due to data fusion challenges
Solution Approach 1:
The patent replaces complex manual data fusion processes with automated natural language processing systems. The NLP technology automatically transforms unstructured soft data into structured semantic data, eliminating the need for manual intervention and reducing system complexity while improving semantic information integration.
Solution Approach 2:
The patent changes the parameter representation of soft data by transforming it from unstructured text form into structured semantic data with defined parameters and relationships through ontology mapping. This parameter transformation enables seamless integration with LiDAR data while managing system complexity through standardized data structures.
3Measurement precision
If semantic labels are associated with point cloud data, then vehicle localization accuracy is improved, but the processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary transformation of soft data into structured semantic data using NLP and ontology before the actual map generation process. This preliminary action prepares the semantic information in advance, allowing for faster integration during localization operations and reducing real-time processing time while maintaining high localization accuracy.
Data Source
AI summary
A method and system for generating a semantic point cloud map. Voice input is received via a microphone and converted into text via speech-to-text synthesis. The text is decomposed into semantic data comprising a number of words, and it is determined whether keywords of the semantic data are present in an autonomous driving (AD) ontology. In response to the keywords of the semantic data being present in the AD ontology, coordinates of a point cloud map corresponding to the semantic data are determined. An association between a semantic label determined from the words of the semantic data and coordinates in the point cloud map corresponding to the semantic data is generated and stored in a memory of the computer vision system.