i.c.sens Multi-view pedestrian tracking dataset

This dataset was acquired for the purpose of collaborative pedestrian tracking from multiple viewpoints. This dataset consists of 29 image sequences which capture trajectories of pedestrians from three different viewpoints with three stereo camera pairs. Each captured image pair(left and right) has a global timestamp. The coressponding image sequences in different views have the same sequence number. The identity number of a pedestrian is unique and identical across viewpoints.

Example tracking results in the first viewpoint and the annotated pedestrians from the other two views: example

The following ".zip" packets are provided for data set usage:

  1. Rectified stereo images captured from three viewpoints, each image has a global timestamp.
  2. The calibration parameters (both intrinsic and extrinsic) of each stereo camera pairs are stored in "calibration.zip" which allows computing the 3D position in the global coordinate of a pixel in image space.
  3. Ground truth annotations of pedestrians appearing in the left image of a stereo are provided for a number of image sequences.

The details of how to use all these data and information are described in the "read_me.txt" of each ".zip" packet

Data and Resources

Cite this as

Uyen Nguyen (2020). Dataset: i.c.sens Multi-view pedestrian tracking dataset. https://doi.org/10.25835/0082741

DOI retrieved: 15:36 01 Oct 2020 (GMT)

Additional Info

Field Value
Author Uyen Nguyen
Last Updated September 14, 2020, 13:14 (CEST)
Created June 23, 2020, 16:53 (CEST)
License Creative Commons Attribution-NonCommercial 3.0