Local Map Fusion of Camera Features and 2D LiDAR for Indoor Mobility
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional SLAM systems face challenges in generating accurate indoor maps due to sensor sensitivity to illuminance and resolution issues, leading to potential collisions and high hardware and software costs, especially when using artificial neural networks.
Innovation Solution
A method and system that binds feature points observed by a camera with 2D LiDAR points and publishes them on a local map, reducing dimensionality and clustering to minimize collisions and computational requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional SLAM uses stereo vision or LiDAR for spatial information, then map generation is achieved, but sensor sensitivity to illuminance and resolution limitations cause incorrect maps and collisions
Solution Approach 1:
The patent combines camera-based visual feature points with LiDAR spatial data to create a hybrid mapping system. This integration allows the system to leverage the high resolution and illuminance insensitivity of LiDAR while incorporating the rich semantic information from camera features, thereby resolving the contradiction between measurement precision and reliability.
Solution Approach 2:
The patent introduces a local map as an intermediary data structure that binds visual feature points with LiDAR points. This local map serves as a mediator that reconciles the differences between visual and spatial data, enabling accurate map generation that overcomes the limitations of individual sensors.
2Measurement precision
If SLAM uses artificial neural network for high accuracy, then mapping precision improves, but hardware costs and computing power requirements increase significantly
Solution Approach 1:
The patent replaces expensive neural network-based SLAM with a more economical approach using traditional computer vision algorithms and local map binding. This solution achieves sufficient mapping precision for indoor mobility without requiring high-end computing hardware, thereby reducing device complexity and cost.
Solution Approach 2:
The patent changes the computational parameters by using dimensionality reduction techniques to convert 3D visual feature points into 2D coordinates that can be directly bound with LiDAR points. This parameter transformation simplifies the computational requirements while maintaining mapping accuracy.
3Loss of information
If LiDAR detects thin legs of furniture, then spatial features are captured, but thin legs are treated as noise and deleted during map generation
Solution Approach 1:
The patent performs preliminary binding of visual feature points with LiDAR points before the map generation and noise filtering stages. By establishing these bindings in advance, the system preserves thin structural features that would otherwise be deleted as noise, while maintaining measurement precision through the bound feature relationships.
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
This approach provides accurate environment data at lower costs without complex computations, reducing collision risks and memory needs, and can be executed on conventional CPUs.
Implementation Method 1
acquiring, by the controller, a 2D LiDAR point data from a LiDAR mounted on the mobility
Implementation Method 2
acquiring, by the controller, a three-dimensional (3D) feature point data by receiving a front image of the mobility from a camera mounted on the mobility
Data Source
AI summary
An embodiment method of generating a local map for travel control of a mobility includes loading the local map from an entire map stored in a memory, wherein the local map includes at least a portion of the entire map, acquiring a two-dimensional (2D) light detection and ranging (LiDAR) point data from a LiDAR mounted on the mobility, acquiring a three-dimensional (3D) feature point data by receiving a front image of the mobility from a camera mounted on the mobility, reducing a dimension of the 3D feature point data to a 2D feature point data, binding the 2D feature point data to the 2D LiDAR point data, and publishing the 2D feature point data bound to the 2D LiDAR point data on the local map.