A profile based data segmentation for in-home activity recognition

Abstract

A major problem in smart-home activity recognition is the ambiguity of interpreting the actions that formulate activities. It resulted from the redundancy of irrelevant actions and the concurrent interleaving among activities themselves. In this paper, we present a framework to minimise the effect of such ambiguity using profile based data segmentation and actions refinement. The proposed methodology relies on defining a profile for each sensor in the environment for enriching existing features with semantic ones. Furthermore, according to these profiles, irrelevant actions within data segments are removed. Moreover, the proposed methodology addresses the interleaving among different activities. Experiments have been conducted to measure the performance of the proposed framework on different datasets. We evaluated our methodology using three different classifiers. The results indicated that the proposed framework achieved statistically significant improvements. Such enhancements resulted from minimising the effect of irrelevant actions and resolving the concurrent interleaving among activities within datasets.

A Novel Feature-Selection Method for Human Activity Recognition in Videos

“…Video activity recognition is a time series classification task that requires combining motion features with video classification models into a machine learning system [16]. This section tracks the progress in activity recognition research, motion representation, and architecture design to handle the learning of spatial and temporal information.…”

“…Electronics 2022, 11, 732 2 of 16 The vast number of features that can be extracted from a video to recognize a human activity complicate the process, making traditional solutions inefficient in terms of several performance indicators such as accuracy, time, and memory. Recently, many attempts have been made to overcome this problem, which adopt several strategies such as Hidden Markov Models [9,10], optical flow-based techniques [11], and most recently, deep-learning [12].…”