Multiple Object Tracking for Dense Pedestrians by Markov Random Field Model with Improvement on Potentials.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Pattern Recognition

Background:

• Pedestrian tracking in dense crowds presents significant challenges for multi-camera systems.
• Existing methods struggle with associating fragmented tracklets caused by occlusions and dense formations.

Purpose of the Study:

• To propose a novel Markov random field (MRF) model for robust association of tracklet couplings in dense pedestrian scenarios.
• To enhance the MRF model with improved potential functions for handling fragmented tracklets.

Main Methods:

• A data fusion method using image mutual information integrates position and motion data for tracklet coupling.
• Human key point detection corrects positional data for incomplete or deviated detections.
• An MRF potential function improvement method incorporates assimilation/extension processing and message selective belief propagation.

Main Results:

• The proposed MRF model effectively associates fragmented tracklet coupling segments in dense crowds.
• Assimilation and extension processing enhance fragmented tracklet information and adjacent node potentials.
• Message selective belief propagation prevents the spread of unreliable messages, improving overall tracking accuracy.

Conclusions:

• The developed MRF model offers a robust solution for pedestrian tracking in challenging dense crowd environments.
• The method demonstrates superior tracking performance validated through modular and system-level experiments on the PETS2009 dataset.
• This research contributes to more reliable crowd analysis and surveillance systems.