SLAM Initialization from Two Camera Poses for Global Convergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing SLAM processes face challenges in accurately mapping and localizing a mobile computing device within a physical environment due to errors in matching corresponding features between captured images, often converging to local minima rather than global minima, especially when using limited computing resources and noisy sensors.
Innovation Solution
The SLAM process is initialized using images from two distinct poses, correlating image blur with movement sensors to detect key frames, and employs a partial SLAM process to reduce resource requirements, allowing convergence to global minima, even with noisy IMUs, and supports augmented reality applications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional SLAM processes use multiple images to improve localization accuracy, then mapping precision improves, but computing resources and time consumption increase
Solution Approach 1:
The patent performs preliminary actions by detecting movement between image frames using IMU sensors before initiating the full SLAM process. By identifying key frames based on movement detection, the system prepares only necessary image pairs for processing, reducing overall computing resource consumption while maintaining localization accuracy.
Solution Approach 2:
The patent extracts and utilizes IMU movement data separately from image processing. By taking out the movement detection function and performing it independently using sensor data, the system identifies candidate image pairs without requiring full image analysis, thereby reducing computing resource requirements for the SLAM process.
2Measurement precision
If SLAM processes use feature matching between multiple images to improve mapping accuracy, then mapping precision improves, but the system converges to local minima rather than global minima
Solution Approach 1:
The patent performs preliminary movement detection using IMU sensors to identify key frames before performing feature matching. This preliminary action provides a reliable indication of actual device movement, allowing the SLAM process to focus on image pairs where global minima convergence is more likely, rather than attempting to match features across all possible image combinations.
Solution Approach 2:
The patent introduces IMU movement detection as an intermediary between image capture and SLAM processing. This intermediary provides reliable movement information that guides the selection of image pairs for feature matching, helping the system avoid local minima by focusing computational effort on pairs with verified device movement between them.
3Measurement precision
If SLAM processes process all captured images to improve localization precision, then measurement precision improves, but processing time and computational load increase
Solution Approach 1:
The patent extracts movement information from IMU sensors and uses it to filter the set of images requiring SLAM processing. By taking out the movement detection function and applying it as a pre-filter, the system processes only the subset of image pairs where device movement occurred, significantly reducing processing time while maintaining localization precision.
Solution Approach 2:
The patent performs partial SLAM processing by applying the full SLAM algorithm only to image pairs identified as containing device movement, rather than processing all captured images. This partial action approach maintains localization precision for movement events while avoiding the time cost of processing static image pairs.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Various embodiments provide systems, methods, devices, and instructions for performing simultaneous localization and mapping (SLAM) that involve initializing a SLAM process using images from as few as two different poses of a camera within a physical environment. Some embodiments may achieve this by disregarding errors in matching corresponding features depicted in image frames captured by an image sensor of a mobile computing device, and by updating the SLAM process in a way that causes the minimization process to converge to global minima rather than fall into a local minimum.