VCHAR: A Variance-Driven Framework for Complex Human Activity Recognition With Vague Label
- 24 November 2025
- journal article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (4) , 5621-5634
Abstract
Complex human activity recognition (CHAR) remains a pivotal challenge within ubiquitous computing, especially in the context of smart environments. Existing studies typically require meticulous labeling of both lower level and higher level activities—a task that is labor-intensive and prone to errors, focusing on datasets that precisely label simple activities or their sequences. These approaches are often impractical in real-world settings due to the scarcity and inaccuracies of available datasets. In response to these challenges, we introduce variance-driven complex human activity recognition (VCHAR), a novel framework capable of detecting both low-level activities and their corresponding high-level activities without requiring precise temporal or sequential labeling of the lower level activities. We introduce a novel approach by applying the KL divergence as the objective function, modeling the outputs of lower level activities as distributions over target time intervals while concurrently detecting higher level activities. This method only requires vaguely labeled activities, removing the need for precise, second-by-second annotations. Additionally, to enhance the layperson's understanding of our model's results, we are the first to apply a novel generative method that explains the result of the sensor model through video, making the information more accessible and understandable. Our evaluation across three publicly available datasets demonstrates that VCHAR uses KL divergence to model the lower level activity distribution over specific time intervals, enhancing the accuracy of complex activity recognition. Furthermore, user studies confirm that VCHAR's explanations are more intelligible compared to existing methods, facilitating a broader understanding of complex activity recognition among nonexperts.Keywords
Funding Information
- National Science Foundation (NSF) under the Center for Smart Streetscapes through NSF Cooperative (EEC-2133516)
This publication has 36 references indexed in Scilit:
- Plug-and-Play Diffusion Features for Text-Driven Image-to-Image Translation2023
- FLAVA: A Foundational Language And Vision Alignment Model2022
- High-Resolution Image Synthesis with Latent Diffusion Models2022
- Human activity recognition by combining external features with accelerometer sensor data using deep learning network modelMultimedia Tools and Applications, 2022
- Deep learning based multimodal complex human activity recognition using wearable devicesApplied Intelligence, 2020
- Human Daily Activity and Fall Recognition Using a Smartphone’s Acceleration Sensor2017
- The Opportunity challenge: A benchmark database for on-body sensor-based activity recognitionPattern Recognition Letters, 2013
- The Mining and Analysis Continuum of Explaining Uncovered2010
- Likert Scale2007
- Activity Recognition from User-Annotated Acceleration Data2004