Map Matching Trajectories Using Pose Graph Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for aligning mobile device trajectories with a map, especially in GNSS-unavailable areas, face inaccuracies due to noisy motion constraints and unreliable loop closures, lacking a mechanism to leverage full physical map constraints.
Innovation Solution
A computer-implemented method that processes a pose graph through non-linear optimization, extracts sub-graphs for map matching, generates new constraints, and iteratively fuses these constraints to refine the alignment, ensuring accurate alignment of mobile device trajectories with a map.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If non-linear optimization is used to process trajectories, then trajectory alignment is achieved, but accuracy deteriorates due to noisy motion constraints and unreliable loop closures
Solution Approach 1:
The patent segments the pose graph into multiple sub-graphs based on spatial proximity and trajectory overlaps. Each sub-graph is processed independently through map matching to generate localized constraints, which are then fused across sub-graphs. This segmentation prevents error propagation across the entire graph and allows more reliable local alignments to inform global trajectory positioning.
Solution Approach 2:
The patent introduces map features as intermediary elements that mediate between trajectories and the global map frame. Instead of directly constraining trajectories to each other through unreliable loop closures, the system matches trajectory sub-graphs to map features (walls, corridors, landmarks), using these map features as stable reference points to establish accurate relative positioning.
2Measurement precision
If map matching is applied to the entire pose graph, then global alignment is achieved, but computational complexity increases significantly
Solution Approach 1:
The patent divides the large pose graph into multiple smaller sub-graphs based on spatial criteria such as proximity of trajectories and overlap regions. Each sub-graph contains a limited number of poses and can be processed independently through map matching, reducing the computational burden compared to processing the entire graph at once. The sub-graphs are then fused together to achieve global alignment.
Solution Approach 2:
Instead of applying map matching to the complete pose graph, the patent applies it selectively to extracted sub-graphs that represent local regions of interest. This partial action approach processes only necessary portions of the data at high resolution, while the overall global alignment emerges from the fusion of these partial results, reducing total computational requirements.
3Reliability
If loop closure constraints are used to align trajectories, then relative positioning is improved, but accuracy deteriorates due to constraint contradictions and noise
Solution Approach 1:
The patent introduces map features as intermediary reference objects that mediate between trajectories and the global coordinate system. Instead of directly enforcing loop closure constraints between trajectory poses (which can be noisy and contradictory), the system matches trajectory sub-graphs to stable map features like walls and corridors. These map features serve as reliable intermediaries that provide consistent geometric references for positioning.
Solution Approach 2:
The patent performs preliminary map matching on trajectory sub-graphs before finalizing loop closure constraints. By first aligning sub-graphs to map features and generating initial constraints from these alignments, the system establishes a more reliable foundation for subsequent optimization steps, reducing the impact of noisy direct loop closure measurements.
Data Source
AI summary
A computer implemented method (400) of locating a plurality of mobile device trajectories (301) relative to each other and a map (1, 1′), the mobile device trajectories (301) comprising a time series of position nodes (303) joined by edges (305), the method (400) comprising: obtaining an input pose graph (311) comprising a plurality of mobile device trajectories (301) and a plurality of first constraints (309) defining the position and orientation of nodes (303) and edges (305) of the trajectories relative to each other; performing a non-linear optimisation process on the input pose graph (311) based on the first constraints (309), to reduce a cost function associated with the first constraints (309), the non-linear optimisation process providing a modified pose graph (313); extracting one or more sub-graphs from the modified pose graph (313), and for each sub-graph, individually processing the sub-graph to map match the nodes (303) and edges (305) of the sub-graph to features defined in the map (1,1′); and generating second constraints (317) based on the one or more map matched sub-graphs.