#Flickr-Photo Extract users trajectories from geo-tagged photo/video data of YFCC100M dataset.

---

Trajectory construction

Trajectories are constructed through following three major steps (More detail descriptions are provided in trajectory_construction.ipynb):

1. Extract photos/videos taken near Melbourne area from original YFCC100M dataset (src/filtering_bigbox.py)
2. Extract initial trajectories based on the extracted photos/videos (src/generate_tables.py)
3. Filter out some abnormal trajectories using various criteria (src/trajectory_construction.ipynb)

---

Trajectory Analysis

Some statistics and analysis about the extracted trajectories are provided in src/flickr_analysis.ipynb

---

Output files

###data Data files

• Melbourne-bbox.kml The two (big/small) bounding boxes of Melbourne used for extracting relevant photos/videos from original YFCC100M dataset
• Melbourne-bigbox.csv Photos/videos taken inside of the big bounding box, output of src/filtering_bigbox.py
• trajectory_photos.csv Trajectorie file, output of src/generate_tables.py
• Each line represents one photo/video with following information
• Trajectory_ID: trajectory ID of entry (multiple entries consist single trajectory)
• Photo_ID: Unique Photo(Video) ID of entry
• User_ID: User ID
• Timestamp: When a photo/video has been taken
• Longitude: Longitude of entry
• Latitude: Latitude of entry
• Accuracy: GPS Accuracy level (16 - the most accurate, 1 - the least accurate)
• Marker: 0 if the entry is photo, 1 if the entry is video
• URL: flickr URL to the entry
• trajectory_stats.csv Stats for each trajectory, output of src/generate_tables.py
• Each line shows statistics about corresponding trajectory
• Trajectory_ID: Unique trajectory ID
• User_ID: User ID
• #Photo: Number of photos+videos in the trajectory
• Start_Time: When the first photo/video has been taken
• Travel_Distance(km): Sum of the distances between consecutive points (Euclidean Distance)
• Total_Time(min): The time gap between the first photo and the last photo
• Average_Speed(km/h): Travel_Distances(km)/Total_Time(h)

src Source files

• filtering_bigbox.py Python3 scripts to extract photos taken in the big bounding box from YFCC100M dataset
• generate_tables.py Python3 scripts to generate trajectories and stats of trajectories
• traj_visualise.py Python3 scripts to generate KML files to visualise trajectories

---

POI and Trajectory data in Melbourne

Notebooks

• src/poi_wikipedia.ipynb IPython notebook to extract POI information from Wikipedia webpages.
• src/traj_Melb.ipynb IPython notebook to construct trajectories by mapping photos to POIs.
• src/dataset_Melb.ipynb IPython notebook to describe the dataset of trajectories in Melbourne.
• src/suburb-name.ipynb Adds the suburb name based on ABS SLA2 to data/poi-Melb-all.csv and saves it to data/poi-Melb-all-suburb.csv GDAL is annoying to install

Data

• data/poi-Melb-all.csv POI data file, generated by IPython notebook src/poi_wikipedia.ipynb, data/poi-Melb-all-suburb.csv contains suburb names of POIs.

• poiID: POI unique ID

• poiName: POI Name

• poiTheme: POI Category

• poiLat: POI Latitude

• poiLon: POI Longitude

• poiURL: URL of Wikipedia webpage that describes this POI

• poiPopularity: The popularity of POI, i.e., the number of distinct users that visited the POI

• data/userVisits-Melb.csv Trajectory data file, generated by IPython notebook src/traj_Melb.ipynb.

• photoID: Flickr Photo unique ID

• userID: User ID

• dateTaken: The date that this photo was taken

• poiID: POI unique ID

• poiTheme: POI Category

• poiFreq: Number of photos taken at the POI

• seqID: Trajectory unique ID

• data/traj-noloop-all-Melb.csv Trajectories extracted from data/userVisits-Melb.csv in an approach that no loops/subtours exist,i.e. use the order of the first occurrences of POIs to form a trajectory.

• userID: User ID

• trajID: Trajectory ID

• poiID: POI ID

• startTime: When a user start to visit the POI, approximated by the time the first photo taken by the user at that POI

• endTime: When a user leave the POI, approximated by the time the last photo taken by the user at that POI

• #photo: Number of photos taken at the POI by the user

• trajLen: Number of POIs in the trajectory

• poiDuration: Visit duration (in seconds) at the POI

• data/Melb_recommendations.csv Trajectory recommendation results of different methods using data/poi-Melb-all.csv and data/traj-noloop-all-Melb.csv. Recommendation results are obtained from each algorithm given the first and last POIs and length of the real trajectory. NA represent failed recommendation due to e.g. ILP timeout. data/Melb_recommendations_F1.csv is the file with F1-scores of the recommended trajectories by different methods and data/Melb_recommendations_pairsF1.csv is the file with pairs-F1-scores.

• trajID: Trajectory ID

• REAL: The ground truth trajectory

• PoiPopularity: Trajectory recommended using POI popularity only

• PoiRank: Trajectory recommended using POI ranking by rankSVM

• Markov: Trajectory recommended using POI-POI transition matrix and Viterbi decoding

• MarkovPath: Trajectory recommended using POI-POI transition matrix and integer linear programming (ILP)

• Rank+Markov: Trajectory recommended using both POI ranking by rankSVM and POI-POI transition matrix, and Viterbi decoding

• Rank+MarkovPath: Trajectory recommended using both POI ranking by rankSVM and POI-POI transition matrix, and ILP

• StructuredSVM: Trajectory recommended using Structured Support Vector Machine

• PersTour: Trajectory recommended using method described in this paper

• PersTour-L: Trajectory recommended by a method similar to PersTour, with time constraint replaced by length constraint

• Visualisation: Please import these KMZ files to Google My Maps to visualise.

• data/Melb_POI.kmz POIs in Melbourne dataset.

• data/Melb_transition_most_popular_POI.kmz Transitions from the most popular POI, i.e., the Federation Square.

• data/Melb_transition_Margaret_Court_Arena.kmz Transitions from the Margaret Court Arena.

• data/Melb_transition_Queen_Victoria_Market.kmz Transitions from the Queen Victoria Market.

• data/Melb_transition_RMIT_city.kmz Transitions from the RMIT city.

• data/Melb_transition_RMIT_city_top30.kmz Top 30 transitions from the RMIT city.

• data/Melb_transition_University_of_Melbourne.kmz Transitions from the University of Melbourne.

• data/Melb_transition_University_of_Melbourne_top30.kmz Top 30 transitions from the University of Melbourne.

• data/Melb_traj_pass_Government_House.kmz Trajectories that include the Government House.

• data/Melb_traj_pass_Melbourne_Cricket_Ground.kmz Trajectories that include the Melbourne Cricket Ground (MCG).

• data/Melb_traj_recommendation_example.kmz Example of trajectories recommended by different approaches.

• data/Melb_terrible_traj_photo_seq.kmz Example of an ugly trajectory.