3D Point Cloud SLAM with IMU Decoupling for MUAV Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing SLAM methods are challenging to implement on Micro Unmanned Aerial Vehicles (MUAV) due to the lack of odometer information in GPS-denied environments and limited computational resources, making it difficult to achieve real-time 3-D dense mapping with sufficient accuracy.
Innovation Solution
A simultaneous localization and mapping method that decouples 3-D point cloud data using inertial measurement data to reduce computational complexity, applying rotation to align point cloud data with key frames and using axonometric projections to estimate translation, and employing a kernelized correlation filter for efficient data association.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If iterative solutions are used in existing SLAM methods, then data association can be performed, but computational complexity increases and processing speed decreases
Solution Approach 1:
The patent segments the 6-DOF pose estimation problem into two independent sub-problems: 3-DOF rotation estimation and 3-DOF translation estimation. This is achieved by decoupling the transformation matrix into rotational and translational components, allowing each to be solved separately. The rotation is estimated using point cloud alignment, while translation is estimated using axonometric image correlation, significantly reducing computational complexity compared to iterative 6-DOF solutions.
Solution Approach 2:
The patent extracts the rotation component from the full pose estimation problem and solves it independently first. By removing the rotational degrees of freedom from the translation estimation problem, the remaining translation can be solved more efficiently using 2D image correlation techniques rather than full 3D point cloud registration.
2Measurement precision
If dense maps are produced for collision-free motion planning and object detection, then mapping precision is improved, but computational resources required increase
Solution Approach 1:
The patent replaces computationally intensive 3D point cloud processing with 2D axonometric image processing for translation estimation. By projecting 3D points onto 2D axonometric images and using correlation filters, the system achieves dense mapping with significantly reduced computational power requirements, making it suitable for embedded systems on micro-robots.
3Measurement precision
If 3-D Mapping algorithms are applied to micro-robot systems, then mapping capability is achieved, but payload and power limitations are exceeded
Solution Approach 1:
The patent uses lightweight axonometric image projections instead of maintaining full 3D point cloud structures for translation estimation. The 2D correlation filters and axonometric images require minimal memory and processing power, enabling dense mapping capabilities on micro-robots with severe payload and power constraints where traditional 3D mapping algorithms would be infeasible.
Data Source
AI summary
A simultaneous localization and mapping method and corresponding system are disclosed. The method comprises: receiving a frame of three dimensional point cloud data from a three dimensional point cloud sensor coupled to a robot; receiving inertial measurement data indicative of a change in pose of the three dimensional point cloud sensor; using the inertial measurement data to estimate a rotation between the received frame of three dimensional point cloud data and a key frame of three dimensional point cloud data; applying a rotation to the received frame of three dimensional point cloud data to obtain an aligned frame of three dimensional point cloud data, the aligned frame of three dimensional point cloud data having an orientation aligned with the key frame of three dimensional point cloud data; and estimating a translation of the aligned frame of three dimensional point cloud data from the key frame of three dimensional point cloud data.