Correspondence-free activity analysis and scene modeling in multiple camera views.

Area of Science:

• Computer Vision
• Machine Learning
• Pattern Recognition

Background:

• Analyzing object activities in synchronized camera networks is challenging due to uncalibrated views and arbitrary camera topologies.
• Existing methods often rely on solving the complex object correspondence problem beforehand.

Purpose of the Study:

• To develop a novel, unsupervised approach for activity analysis in multi-camera systems.
• To group object trajectories belonging to the same activity across different views.
• To model common object paths and detect anomalous activities.

Main Methods:

• Objects are tracked independently in each camera view, extracting position and velocity features.
• A probabilistic model jointly learns activity distributions across feature spaces of different camera views.
• Unsupervised learning is employed, eliminating the need for pre-established object correspondence.

Main Results:

• The approach successfully groups trajectories from different views into activity clusters.
• It models common object movement patterns across multiple camera perspectives.
• The method demonstrates effectiveness in detecting abnormal activities.
• Evaluation on simulated and large real-world datasets (22,951 and 14,985 trajectories) validates the approach.

Conclusions:

• This novel method provides an effective unsupervised solution for multi-camera activity analysis.
• It bypasses the need for solving the correspondence problem, simplifying the analysis.
• Learned activity models can subsequently aid in solving the correspondence problem.