Plane Extraction for Drift-Resistant Device Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing localization techniques for mobile devices, such as smartphones, are inefficient and consume high power due to reliance on sensors like IMUs and GPS, which fail to provide accurate and efficient localization in real-time environments, especially indoors, and struggle with environments lacking texture.
Innovation Solution
A method involving detecting and extracting planes from depth data using a two-stage process to correct IMU data, where the first stage detects likely planes and the second stage refines them for precise parameter estimation, enhancing localization accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Duration of action of moving object
If IMU sensors are used for localization, then device position and orientation can be tracked, but the estimates diverge over time due to sensor noise and bias
Solution Approach 1:
The system uses detected planes as feedback references to correct IMU integration drift. By continuously comparing IMU-derived pose estimates against plane constraints (normal vectors and distances), the system periodically corrects accumulated errors, enabling long-duration accurate localization.
Solution Approach 2:
Planes serve as intermediary reference objects that mediate between IMU sensors and the environment. The planes provide stable geometric constraints that bridge the gap between inertial measurements and physical space, enabling accurate pose estimation without direct GPS or visual features.
2Measurement precision
If point features are extracted from images for VIO, then additional motion information can be obtained, but the scene may contain very few point features due to lack of texture
Solution Approach 1:
The system changes the type of geometric features used for localization from point features to plane features. Planes provide stable constraints (normal vectors and distances) that work reliably in textureless environments where point features fail, significantly improving environment adaptability.
Solution Approach 2:
The system transitions from 3D point feature tracking to plane parameter estimation, adding the dimension of surface orientation (normal vectors) and distance constraints. This dimensional shift enables robust localization in environments lacking texture by utilizing the geometric properties of planar surfaces.
3Measurement precision
If a two-stage plane detection and extraction process is used, then precise plane-parameter estimation can be achieved, but computation time increases
Solution Approach 1:
The plane detection process is segmented into two stages: initial detection using RANSAC for rapid plane hypothesis generation, followed by refinement using orthogonal regression for precise parameter estimation. This segmentation allows the system to achieve high precision while managing computation time by performing coarse detection first, then refining only necessary candidates.
Solution Approach 2:
The system performs preliminary plane detection using computationally efficient RANSAC before conducting more intensive refinement. This preliminary action filters out incorrect plane hypotheses early, reducing the computational burden of subsequent precise parameter estimation while maintaining accuracy.
Data Source
AI summary
Various implementations disclosed herein include devices, systems, and methods that localize a device based on detecting planes in depth data acquired by the device. For example, an example process may include detecting first plane data in first sensor data acquired by a sensor at a first viewpoint location in a physical environment, detecting second plane data in second sensor data acquired by the sensor at a second viewpoint location in the physical environment, determining that the first plane data and the second plane data correspond to a same plane based on comparing the first plane data with the second plane data, and determining a spatial transformation between the first viewpoint location and the second viewpoint location based on the first plane data and the second plane data.