This paper presents a unified framework that jointly optimizes indoor user localization and human activity recognition (HAR) through the Alternating Direction Method of Multipliers (ADMM). By exploiting the inherent spatial-activity coupling in human behavior, the method decomposes the joint optimization into localized subproblems solved iteratively with guaranteed convergence. The problem is formulated as a multi-task learning objective with structural regularization, enabling information sharing between localization and activity detection modules while maintaining computational efficiency. Extensive experiments on benchmark datasets demonstrate superior performance compared to independent baselines.
Key findings
Joint optimization of localization and activity detection through ADMM improves performance.
Exploiting spatial-activity coupling in human behavior leads to better optimization.
The proposed method achieves 15% improvement in localization accuracy and 12% in activity recognition F1-score.
Limitations & open questions
The framework's performance in real-time deployment needs further validation.
The complexity of the relationship between location and activity in different contexts is not fully explored.