event_geolocalization_dataset

Getting Started

Please refer following paper for details on the experiments in this dataset:
"A Pseudo-likelihood Approach For Geo-localization of Events From Crowd-sourced Sensor-Metadata".

The metadata corresponding to the ith experiment can be found in "dataset/exp_i.pk".
The information about the video files can be found in "video_info/exp_i.pk".
The individual files can be read using python2.7 pickle syntax as shown in read_data.py and read_video_info.py, respectively.

Sensor Metadata

The file "dataset/exp_i.pk" contains dictionary with following fields:

• geovid_server_query_parameters:

  its a tuple containing parameters (start_date,start_time,end_date,end_time,SW_lat,SW_lng,NE_lat,NE_lng,location_timezone_UTC) used for searching the videos of the experiments uploaded on the geovid.org server.
  The tuple contains data (start_date, start_time, end_date, end_time) used for defining temporal bound for searching, (SW_lat, SW_lng, NE_lat, NE_lng), i.e. south-west lattitude and longitude, and north-east lattitude and longitude used for defining geographic bounds, and location_timezone_UTC time zone of the locations where experiments were carried out.
  Please refer "Sakire Arslan Ay, Lingyan Zhang, Seon Ho Kim, Ma He, and Roger Zimmermann. 2009. GRVS: A georeferenced video search engine. In ACM Proc. Multimedia. 977–978" for more details on how to use these parameters for creating a query url and searching the videos and sensor data.

• is_event_track_gps_available:
  this field indicates whether gps data for the event track is available.

• number_of_cameras:
  number of cameras, N, video recording the event.

• number_of_timestamps:
  total number of sensor metadata samples, T, taken during the experiment. Sampling period is 200ms.

• mean_coords_in_meters:
  the tuple (\mu_x, \mu_y), these coordinates can be added into the camera/event coordinates to get corresponding coordinates on the surface of the earth (in meters).

• sensor_data:
  tuple containing (camera_x_coordinate, camera_y_coordinate, camera_orientation, camera_magnetometer_strength, camera_gps_error_variance). Each variable is a 2D array of size TxN such that (t,n) entry represents data for n^{th} camera for time instant t.

• event_track:
  this field is valid if is_event_track_gps_available is True. It contains tuple (event_x_coordinate, event_y_coordinate, event_gps_error). Each variable is 1D array of size T such that t^{th} entry represents data for time instant t representing gps data for event location and corresponding gps error.

• raw_data:
  dictionary with following fields: 'gps_data', 'compass_data', 'accelerometer_data'

  • gps_data:
    raw gps sensor measurements during the experiment in following format [gps_data_cam_0, gps_data_cam_1, ..., gps_data_cam_(N-2), gps_data_cam_(N-1), gps_data_event] if "is_event_track_gps_available" is True else [gps_data_cam_0, gps_data_cam_1, ..., gps_data_cam_(N-2), gps_data_cam_(N-1)] each field in above list is as below gps_data_cam_x: [data(0), data(1), ... ] where data(t) is [UTC timestamp since unix epoch, Latitude, Longitude, GPSError_std, Altitude] Note: sampling frequency for gps data is 1Hz.

  • compass_data:
    raw compass sensor measurements during the experiment in following format [compass_data_cam_0, compass_data_cam_1, ..., compass_data_cam_(N-2), compass_data_cam_(N-1), compass_data_event] if "is_event_track_gps_available" is True else [compass_data_cam_0, compass_data_cam_1, ..., compass_data_cam_(N-2), compass_data_cam_(N-1)] each field in above list is as below compass_data_cam_x: [data(0), data(1), ... ,data(T-2),data(T-1)] where data(t) is [UTC timestamp since unix epoch, hx, hy, hz, (hx*hx, hy*hy, hz*hz)**.5] where (hx,hy,hz) is a magnetometer measurement.

  • accelerometer_data:
    raw accelerometer sensor measurements during the experiment in following format [accelerometer_data_cam_0, accelerometer_data_cam_1, ..., accelerometer_data_cam_(N-2), accelerometer_data_cam_(N-1), accelerometer_data_event] if "is_event_track_gps_available" is True else [accelerometer_data_cam_0, accelerometer_data_cam_1, ..., accelerometer_data_cam_(N-2), accelerometer_data_cam_(N-1)] each field in above list is as below accelerometer_data_cam_x: [data(0), data(1), ... ,data(T-2),data(T-1)] where data(t) is [UTC timestamp since unix epoch, ax, ay, az, (ax*ax, ay*ay, az*az)**.5] where (ax,ay,az) is a magnetometer measurement.

Video Data

The file "video_info/exp_i.pk" contains a dictionary with information about video recordings. Each entry in the dictionary is a list of length N (or N+1 if is_event_track_gps_available is True). The video recordings are also available to download.

• start_timestamps:
  a list containing sequence of UTC timestamp since unix epoch indicating time when the individual video recordings started.

• stop_timestamps:
  a list containing sequence of UTC timestamp since unix epoch indicating time when the individual video recordings ended.

• video_durations:
  a list containing sequence of time durations in milliseconds indicating durations of individual video recordings.

• start_offsets:
  a list containing sequence of offset values in milliseconds with respect to start_timestamps indicating when the event video recordings started.

• stop_offsets:
  a list containing sequence of offset values in milliseconds with respect to stop_timestamps indicating when the event video recordings ended.

• event_recording_durations:
  a list containing sequence of time durations in milliseconds indicating durations of individual event video recordings.

Please refer read_video_info.py for more details. Whenever an experiment have is_event_track_gps_available to be True, the last entries in the above lists corresponds to a video camera moving along with the event.

Authors

• Amit More - amitmore17

contact:[email protected],[email protected].
Please send an email on both email addresses for quick reply.