Pedestrian Map Inference Using Kernel Density Estimation and Trace Merging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current map inference techniques are inadequate for generating accurate and complete pedestrian navigable maps due to low-quality geolocation measurements from mobile devices, which are noisy and have low sampling rates, differing significantly from vehicle-based GPS data.
Innovation Solution
A three-step framework integrating kernel density estimation (KDE) and trace merge algorithms, utilizing starting and destination points to extract high-quality walking traces, even from sparse data, and converting these into navigable map data, robust against noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If kernel density estimation (KDE) is used to handle satnav noise, then noise robustness is improved, but it requires very dense sampling points which are not available in crowd-sourced pedestrian satnav measurements
Solution Approach 1:
The patent combines KDE with trace merge algorithms to leverage the noise robustness of KDE while using trace merging to achieve sufficient sampling density by aggregating multiple pedestrian trajectories. This hybrid approach allows the system to handle noisy crowd-sourced satnav measurements effectively without requiring individually dense sampling from each device.
Solution Approach 2:
The patent performs preliminary clustering and trace merging before applying KDE, thereby pre-concentrating the sampling points along inferred paths. This preliminary action creates the effective density needed for KDE to work properly, even when individual pedestrian trajectories are sparse.
2Measurement precision
If trace merge algorithm is used to cluster nearby traces, then clear trace extraction is improved, but it requires high quality satnav measurements with sufficiently dense traces which pedestrian mobile phone data does not provide
Solution Approach 1:
The patent adapts the trace merge algorithm by modifying its parameters and thresholds to work with lower quality pedestrian satnav data. Instead of requiring the high precision and density needed for vehicle GPS, the algorithm uses adjusted clustering parameters that are suitable for noisier, sparser mobile phone measurements, thereby extracting usable traces from degraded data quality.
3Productivity
If k-means technique is used to cluster nearby pings, then map inference efficiency is improved, but it requires clear and sufficiently dense traces which are not available from low-quality pedestrian satnav measurements
Solution Approach 1:
The patent segments the map inference process into multiple stages: initial clustering of raw pings using k-means, followed by trace merging, and finally KDE-based refinement. This segmentation allows each stage to handle specific aspects of the data quality problem, with later stages compensating for the limitations of earlier stages when dealing with sparse pedestrian data.
4Area of stationary object
If crowd-sourced satnav measurements from mobile phones are used, then data collection coverage is improved, but measurement accuracy and sampling rate deteriorate compared to vehicle-based GPS
Solution Approach 1:
The system uses the large volume of crowd-sourced pedestrian measurements to compensate for individual measurement quality. By aggregating data from many users, the system achieves sufficient statistical density and coverage, allowing the lower-accuracy individual measurements to collectively produce high-quality map inference results.
Data Source
AI summary
Communications server apparatus (102) for generating navigable map data comprises a processor (116) and a memory (118), and is configured, under control of the processor, to execute instructions (120) stored in the memory: to generate a series (340) of journey trace data sets (338), each journey trace data set comprising data representing a user journey (310), each journey trace data set being derived using geolocation transmissions (318) from a communications device (104, 106) of a user (308) undertaking the user journey; to aggregate the series of journey trace data sets to generate route image data comprising data representing a network of navigable routes; and to generate, from the route image data, the navigable map data, the navigable map data comprising data representing a series of geolocations corresponding to the network of navigable routes.