Related Experiment Video
Updated: Jun 8, 2026
08:25
Combining Eye-tracking Data with an Analysis of Video Content from Free-viewing a Video of a Walk in an Urban Park Environment
Published on: May 7, 2019
Spatiotemporal localization and categorization of human actions in unsegmented image sequences
Antonios Oikonomopoulos1, Ioannis Patras, Maja Pantic
1Department of Computing, Imperial College London, London SW7 2AZ, UK. aoikonom@doc.ic.ac.uk
Summary
This study introduces a novel method for human activity recognition and localization in videos. It effectively identifies activities even with clutter and occlusion using a probabilistic voting scheme.
Area of Science:
- Computer Vision
- Machine Learning
- Human Activity Recognition
Background:
- Human activity recognition in unsegmented image sequences is challenging.
- Existing methods struggle with clutter, occlusion, and multiple simultaneous activities.
Purpose of the Study:
- To develop a robust method for localizing and recognizing human activities in unsegmented image sequences.
- To handle complex scenarios including clutter, occlusion, and multiple activities.
Main Methods:
- Utilizes an implicit representation of spatiotemporal activity shape via feature descriptor ensembles.
- Employs a probabilistic spatiotemporal voting scheme for evidence accumulation.
- Uses boosting to select class-specific codebooks of feature ensembles.
Main Results:
- Demonstrates effective localization and recognition of human activities.
- Successfully handles multiple activities, clutter, and occlusion in image sequences.
- Achieves accurate results on publicly available datasets and challenging custom sequences.
Conclusions:
- The proposed method provides a robust framework for human activity analysis in unsegmented videos.
- The implicit spatiotemporal shape representation and voting scheme are key to handling complex scenes.
- The approach shows significant potential for real-world applications requiring reliable activity recognition.
