Multi-Feature Stationary Foreground Detection for Crowded Video-Surveillance

 

[Description][Related publication][Abstract][Overview][Software][Dataset][Algorithm results]

Description

This page provides the software, ground-truth data and results achieved by the proposed algorithm for Stationary Foreground Detection in crowds. Files with annotations of stationary objects, persons and groups are provided.


Related publication

D. Ortego and J.C. SanMiguel
Multi-Feature Stationary Foreground Detection for Crowded Video-Surveillance
21st IEEE International Conference on Image Processing (ICIP 2014), Paris (France), (Accepted)

Contact Information
Diego Ortego -
show email


Abstract

We propose a novel approach for stationary foreground detection in crowds based on the spatio-temporal evolution of multiple features. A generic framework is presented to detect stationarity where history images model the spatio-temporal feature patterns. A feature is proposed based on structural information over each pixel neighborhood for dealing with shadows and illumination changes. A multifeature detector is composed by combining the history images of three features (namely, foreground, motion and structural information) to estimate the foreground stationarity over time, which is later thresholded to detect stationary regions. Experimental results over challenging video-surveillance sequences show the improvement of the proposed approach against related work as structural information reduces false detections, which are common in crowded places.


Overview

In our proposed approach N=3, thus computing three History Images (HI). Features extracted are: Foreground (FG), Motion (MO) and Structure (ST). HI combination yields to an image that describes spatio-temporal stationarity (HIS). Thresholding the mentioned combination a Static Mask (SFGt), robust to illumination changes and continuous motion areas, is obtained.


Software

Multi-Feature Stationary Foreground Detection software is available here

Poster from ICIP 2014 is available here


Dataset

Sequences from AVSS 2007 dataset: AVSS07_AB_Easy, AVSS07_AB_Medium, AVSS07_AB_Hard, AVSS07_AB_Eval and AVSS07_PV_Eval. Extracted from here

Sequences from PETS 2006 dataset: PETS06_S1_C1, PETS06_S1_C4, PETS06_S4_C1, PETS06_S4_C2, PETS06_S4_C3, PETS06_S4_C4, PETS06_S7_C1, PETS06_S7_C3 and PETS06_S7. Extracted from here

Sequences from PETS 2007 dataset: PETS07_S5_C1, PETS07_S5_C2 and PETS07_S5_C3. Extracted from here

The ground-truth for all the targets in the dataset is available here


Algorithm results

Results from AVSS 2007 dataset are available here

Results from PETS 2006 dataset are available here

Results from PETS 2007 dataset are available here


Video Processing & Understanding Lab (http://www-vpu.eps.uam.es/)