My Research‎ > ‎

Cerema Metro Station Dataset (CEMEST) for Analyzing and Monitoring Passenger Flow in Public Transport System

Huy Hieu Pham(1,2), Houssam Salmane (2) , Louahdi Khoudour (1) , Alain Crouzil (2) , Pablo Zegers (3) , Sergio A. Velastin (4)

(1) Cerema Research Center, 1 Avenue du Colonel Roche, 31400, Toulouse, France
(2) Informatics Research Institute of Toulouse, Paul Sabatier University, F-31062 Cedex 9, Toulouse, France
 (3) Aparnix, La Gioconda 4355, 10B, Las Condes, Santiago, Chile
 (4) School of Electronic Engineering and Computer Science, Queen Mary University of London, E1 4NS, London, UK 


The Cerema Metro Station Dataset (CEMEST) is an RGB-D dataset for analyzing and monitoring passenger flow in the public transport system. It consists of 203 video samples containing RGB videos, depth map sequences, and 3D skeletal data for each sample. This dataset was captured by Microsoft Kinect v2 sensor and its SDK at a metro station in Toulouse, France. The resolution of RGB videos are 1920×1080 and 3D skeletal data contains the three-dimensional locations of 25 major body joints for each frame (see Figure 2). It has three actions including both normal and abnormal behaviors at the metro station: crossing normally over the barriers, jumping over the ticket barriers, and sneaking under ticket barriersThese behaviors are selected for acquiring because they have a significant impact on monitoring and management in public transport system. The following pictures (Figure 1 (a-i)) show some samples from our dataset.

(a)                                                                            (b)                                                                      (c)
Figure 1: (a-c) Crossing normally over the barriers
(d)                                                                            (e)                                                                      (f)
Figure 1: (d-f) Jumping over the ticket barriers without tickets.
(g)                                                                            (h)                                                                      (i)
Figure 1(g-i) Sneaking under the ticket barriers without tickets.

    Figure 2: Illustration of the human skeleton graphs provided by Microsoft Kinect v2. The picture was reproduced from Microsoft Corporation.

Figure 3 describes our setting for collecting skeleton sequences and storing them as text files for processing later.

Figure 3 Illustration of a skeleton sequence provided by depth sensors. We developed a tool allowing users to record and store the 3D joint coordinates of skeletal data as text files, where each skeleton sequence containing N frames will be converted into a single text file as the structure above. In our study, we set N equal to 16. There are a total of 203 text files that were created.

This package contains the CEMEST dataset version 1.0. We plan to extend our dataset with more action categories, which have a significant impact on public safety such as fighting, explosion, stealing, accident, arrest, etc. with multiple viewpoints.


This dataset was originally recorded for an academic project, and it must be used only for the purposes of research. The skeletal data can be downloaded directly here. For RGB and Depth sequences, please contact Huy Hieu Pham via <>.



Huy Hieu Pham, Houssam Salmane, Louahdi Khoudour, Alain Crouzil, Pablo Zegers, and Sergio Velastin "From skeleton sequences to enhanced motion maps: a new representation for 3D human action recognition with densely connected convolutional networks",  Pattern Recognition LettersSpecial issue on “DLVA: Advances in Deep Learning and Visual Analytics for Intelligent Surveillance Systems" (submitted). 


(will be available soon)


The dataset was created with the collaboration of the Cerema Research Center and the Tisseo network (a public transportation network for Toulouse city, France – to improve security and safety in public transport. We would like to thank you for the support of all our colleagues at Cerema.