A Background Estimation dataset

Presentation


These pages describe a Background Estimation dataset (BEds), a corpus of video sequences generated from publicly available video-surveillance datasets to cover several Background Estimation challenges. 

The dataset is focused on 4 challenges or categories conformed by 10 video sequences each and the associated ground-truth background image. 

Video sequences have been extracted from the following public datasets related with the video-surveillance task:

1) MALL: Video sequences used for crowd counting and profiling research.

2) Bank Street: Video sequences used for anomaly detection.

3) CAVIAR: Video sequences used for event detection.

4) CUHK Crowd Dataset: Video sequences used for Scene-Independent Group Profiling in Crowd.

5) I2R: Video sequences used for Background Subtraction.

6) ATON: Video sequences used for Shadow detection.

7) CVRL: Video sequences used for video-surveillance in indoor scenarios.

8) IITK: Video sequences used for object-tracking, activity recognition, congestion analysis or vehicle counting.

9) CDNET2014: Video sequences used for Background Subtraction.

10) APIDIS: Video sequences used for detection and tracking.

11) AVSS2007: Video sequences used for abandoned object detection an illegally parked vehicle detection.

12) PBI: Video sequences recorded for Background Estimation.

13) Wallflower: Video sequences used for Background Subtraction

14) PETS2009: Video sequences used for pedestrian detection and tracking.

15) LOST: Video sequences recorded for long term observation of scenarios.

16) LIMU: Video sequences used for Background Subtraction.

 For more details please refer to the "Content" section.

Related works (using dataset) :

[1] Carolina Fernández-Pedraza, “Reconstrucción de fondo de escena a partir de secuencias de vídeo&rdquo, Graduate thesis of Grado en Ingeniería de Tecnologías y Servicios de Telecomunicación at Escuela Politécnica Superior (Universidad Autónoma de Madrid). Supervised by Diego Ortego and José M. Martínez.

Work partially supported by the Spanish Goverment under project TEC2014-53176-R (HAVideo).