Virtual Camera Map Optimization for Inside-Out Tracking Drift
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Inside-out tracking in virtual reality systems suffers from map drift due to errors accumulated during map generation, leading to imprecise positioning and virtual environment rendering.
Innovation Solution
A map optimizing method that involves identifying markers in a map, generating virtual cameras with intersecting optical axes based on real-world distances, and performing global bundle adjustment to reduce re-projection errors, thereby optimizing the map and improving positioning accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If inside-out tracking with visual SLAM is used to build a point cloud map, then the system can achieve autonomous positioning and environment mapping without external sensors, but the map is prone to drift due to accumulated errors during map generation
Solution Approach 1:
The patent introduces virtual cameras as an intermediary element to enforce geometric constraints during map optimization. These virtual cameras act as mediators between the real camera observations and the point cloud map, ensuring that the relative positions of markers satisfy the known baseline distance constraints. This intermediary mechanism allows the system to maintain autonomous positioning while correcting accumulated drift errors through constraint-based optimization.
Solution Approach 2:
The patent changes the optimization parameters by incorporating distance constraints between markers as additional optimization variables. Instead of only optimizing based on visual observations, the system now optimizes map points and camera poses while simultaneously satisfying the known baseline distance parameters between multiple markers. This parameter expansion allows the system to correct drift by enforcing geometric consistency across the entire map.
2Reliability
If multiple virtual cameras with intersecting optical axes are generated based on real-world distance constraints, then the system can enforce geometric constraints to reduce map drift, but the computational complexity of global bundle adjustment increases
Solution Approach 1:
The patent segments the optimization problem by processing markers and their associated virtual cameras in discrete groups. Each marker generates a set of virtual cameras with specific geometric constraints, and the optimization process can be broken down into smaller sub-problems centered around each marker-cameras group. This segmentation reduces the overall computational burden compared to treating all cameras and constraints simultaneously as a single large optimization problem.
Solution Approach 2:
The patent applies partial optimization by focusing computational resources on optimizing map points and poses that are constrained by multiple markers, rather than uniformly optimizing all elements in the map. The virtual cameras generated from markers with known baseline distances provide selective constraint enforcement, applying optimization pressure only where geometric constraints exist, thereby reducing unnecessary computational overhead in unconstrained regions of the map.
Data Source
AI summary
A map optimizing method, applicable to an electronic device storing distance values and a map, includes: identifying, from the map, a first map point with a first estimated coordinate and a second map point with a second estimated coordinate generated based on a first marker and a second marker, respectively; generating, according to the first and second estimated coordinates, virtual cameras controlled by a virtual pose and with optical axes intersected at first and second intersection coordinates separated from each other by one of the distance values, in which the virtual cameras provide virtual frames indicating that the first and second markers are observed at the first and second intersection coordinates, respectively; and performing a global bundle adjustment to optimize the map, including adjusting the first and second estimated coordinates to reduce a sum of re-projection errors calculated according to real keyframes and the virtual frames.