Learning to Generate Maps from Trajectories Sijie Ruan , Cheng Long, Jie Bao, Chunyang Li, Zisheng Yu, Ruiyuan Li, Yuxuan Liang, Tianfu He, Yu Zheng sjruan@stu.xidian.edu.cn
Increasing Demands of Accurate & Updated Maps Navigation Route Planning & Real-time Scheduling
Traditional Map Data Collecting Methods Low Spatial Coverage On-field Study Dynamic Traffic Status
Crowdsourcing GPS Trajectories GPS Devices Massive Moving Objects Real-time City-wide Map Information
Challenges Positioning Error (~15 meters) Difficult to Distinguish Spatially Close Roads Low Sampling Rate (30s/pt) Ground Truth Map Ambiguous Underlying Traversing Routes
Framework of DeepMG
Geometry Translation (1/2) β’ Feature Extraction β’ For each π½ Γ πΎ Region Tile 2ΓπΓπΓπ½ΓπΎ ) Spatial View ( β 11Γπ½ΓπΎ ) Transition View ( β€ 2 1 1 0 0 β¦ neighborhood c 0 0 1 0 (b) Local Incoming Matrix. (c) Local Outgoing Matrix. (a) Transitions S/E at c. Point, Line, Speed, Direction (8 channels) Each grid cell has an incoming and an outgoing matrix
Geometry Translation (2/2) β’ T2RNet β’ Transition Embedding β’ Shared Encoder β’ Road Region Decoder β’ Road Centerline Decoder β’ Optimization β’ Dice Loss [Milletari F, et al. 2016] β’ Multi-task Loss Milletari F, et al. β V-net: Fully convolutional neural networks for volumetric medical image segmentation β. 3DV. 2016.
Topology Construction (1/2) β’ Graph Extraction β’ Merge predicted tiles β’ Extract road segments β’ Link Generation β’ For each dead end π 2 π 1 β’ Case 1: intersects another edge on the extension π π β’ Case 2: has smooth transition to the closest dead end of another edge π 3
Topology Construction (2/2) β’ Map Refinement β’ Perform trajectory map matching on the linked map [Yuan J, et al. 2010] β’ Remove edges and links with low support If the map matching is directly appliedβ¦ Proposed solution Inferred path π 1 Real path π 1 π 1 π 3 π½ > 1 π 2 π 2 π 3 π 2 Yuan J, et al. β An Interactive Voting-based Map Matching Algorithm β. MDM. 2010.
Evaluation β’ Datasets β’ Evaluation Metrics β’ Trajectory β’ Topological F1 [Biagioni, et al. 2012] β’ <oid, timestamp, latitude, longitude> β’ Repeat π times β’ Map a. Select a random starting location β’ Node: <latitude, longitude> b. Find reachable area within a maximum radius c. Compare generated map with GT using F1 β’ Edge: <start_node, end_node> β’ Report the average F1 score Biagioni J, Eriksson J. β Inferring road maps from global positioning system traces: Survey and comparative evaluation β. Transportation research record. 2012.
Results β’ Quantitative Comparison TaxiJN (6.5% improvement) TaxiBJ (32.3% improvement) β’ Visual Comparison
Practice in JD.COM: Resident Area Map Generation from JD Couriersβ Trajectories X
Conclusion β’ A valuable but challenging task β’ Position errors β’ Low sampling rate β’ Our method β’ Geometry translation (T2RNet) β’ Convolutional network: learns the structure of the road network β’ Auxiliary task: helps the centerline inference β’ Topology construction (Link + Prune) β’ Trajectories as transition evidences β’ Results β’ Superior than traditional methods β’ More effective on low-sampling rate datasets
Thanks! JD iCity JD Urban Spatio-Temporal Data Engine (JUST) Ruan S., et al. β Learning to Generate Maps from Trajectories β. AAAI. 2020.
Recommend
More recommend
