Smart Route Recommendations based on Historical GPS Trajectories and Weather Information

Transcription

1 Smart Route Recommendations based on Historical GPS Trajectories and Weather Information Xingzhe Xie1, Kevin Wong2, Hamid Aghajan2 and Wilfried Philips1 1 TELIN IPI iminds, Ghent University, Belgium 2 Electrical Engineering, Stanford University, US 1

2 Overview Introduction Approach Overview Intersections detection Features extraction Route planning Results 2

3 Overview Introduction Approach Overview Trajectory segmentation Features extraction Route planning Results 3

4 About my research Room layout exploring and behavior analysis from people's trajectories in smart environments Cameras setup in a smart meeting room Initial image Trajectories Person 1 Person 2 Person 3

5 About my research Initial image Trajectories Person 1 Occupancy map for sitting space Chair locations Person 2 Person 3 Occupancy map for walking space Table location

6 Trajectories from a multi-camera tracking system Smart meeting room GPS tracks Beijing city

7 Aim: find the optimal route between two places

8 Our method with consideration of Traditional method: the shortest geographic route the prior users' experience environmental factors

9 Datasets OpenStreetMap (OSM): A collaborative mapping data contain many types of GIS data including: Road locations and names Points of Interest Natural Features Bodies of Water Political boundaries...

10 Datasets GeoLife Dataset from Microsoft Research Asia: taxi car

11 Datasets Weather information in Beijing from 27 to 212 #Year Day Solar T-max T-min Humidity Precipitation Evaporation

12 Overview Introduction Approach Overview Intersections detection Features extraction Route planning Results OpenStreetMap (OSM) 12

13 Intersections Detection OpenStreetMap (OSM) Road 1: Road 2: Node1 Node5 Node2 Node7 Node3 Node8 Node4 Node9 Node5 Node1 Node6 Node

14 calculate the shortest route based on the connectivity between each pair of nodes, and the distance between them.

15 Connectivity calculation OpenStreetMap (OSM) Road 1: Road 2: Road 3: Node1 Node5 Node78 Node2 Node7 Node89 Node3 Node8 Node4 Node9 Node9 Node5 Node1 Node2 Node6 Node11 Node sum Road 1 Road 3 Node5 Node45 Node9 Road 2

16 Connectivity matrix Node1 Node2 Node3 Node4 Node5 Node6 Node1 1 1 Node Node3 1 Node4 1 1 Node5 1 Node

17 Distance matrix Node1 Node2 Node3 Node4 Node5 Node6 Node1 5m 7m Node2 5m 45m 23m Node3 45m Node4 23m 1m Node5 7m Node6 1m......

18 Overview Introduction Approach Overview Intersections detection Features extraction Route planning Results GeoLife Dataset 18

19 Feature extraction Node5 Features for each pair of connected nodes: average of the speeds standard deviation of the speeds the confidence of the Geolife trajectory data matching the openstreetmap data ave_speed var_speed Node9 GeoLife Dataset track_id Track 1 of user 1 1 Track 45 of user 1 2 Track 6 of user 5 3 Track 3 of user 12 4 confidence of map matching track_id =1 track_id =2 track_id =3 track_id =4 Features of {node5, node9} Precipit ation

20 Map matching OpenStreetMap (OSM) GeoLife Dataset confidence of map matching ave_dist The corresponding segment One segment node i node i+1 node i+2... dist dist1 dist2.. node j node j+1 node j+2...

21 Road evaluation Variables: Variables: Variables: ave_s: average of speeds re: road evaluation var_s: variance of speeds re: road evaluation ave_dist: map matching re: road evaluation

22 Bayesian theory: prob(re) Precipitation track_id =1 track_id =2 track_id =3 track_id =4 corr(road condition, humidity)

23 Overview Introduction Approach Overview Intersections detection Features extraction Route planning Results 23

24 Road evaluation and the impactor of Precipitation are used to adjust the distance between each pair of connected nodes Calculate the shortest route using the connectivity matrix and new distance matrix

25 Overview Introduction Approach Overview Intersections detection Features extraction Route planning Results 25

26 Results intersections

27 Connectivity matrix

28 Road evaluation bad good

29 Weather impactor correlation low high

30 Route planner (traditional method)

31 Route planner (our method)

32 THANKS FOR YOUR ATTENTION! QUESTIONS?