A Dynamic Sliding Window Approach for Activity Recognition

Abstract

Human activity recognition aims to infer the actions of one or more persons from a set of observations captured by sensors. Usually, this is performed by following a fixed length sliding window approach for the features extraction where two parameters have to be fixed: the size of the window and the shift. In this paper we propose a different approach using dynamic windows based on events. Our approach adjusts dynamically the window size and the shift at every step. Using our approach we have generated a model to compare both approaches. Experiments with public datasets show that our method, employing simpler models, is able to accurately recognize the activities, using fewer instances, and obtains better results than the approaches used by the datasets authors.

Dynamic sensor data segmentation for real-time knowledge-driven activity recognition

“…In the data-driven activity recognition community, the problem of sensor data segmentation has been widely explored [1,12,29,[35][36][37]. The notion of time windows is adopted to provide a basis for handling time-dependent data, e.g., the sensor data stream.…”

“…The notion of time windows is adopted to provide a basis for handling time-dependent data, e.g., the sensor data stream. However, some sensor data segmentation approaches use static sliding windows to segment the data stream [1,12,35] while others use dynamically derived time window lengths [29,36,37]. The notion of time slices is used in [35] to derive segments used to perform activity recognition.…”

“…The main criticism for static sliding windows is that incorrect lengths can truncate an activity instance or overlap activity instances leading to recognition failure. Due to the above problem, our work is closely related to [29,36,37] whereby time window parameters are varied.…”

“…Although our work is in the area of knowledge driven activity recognition, it is slightly similar to the work in [37] since it uses dynamic windows. In [37], the length of the window is dynamically derived at runtime based on the occurrence of specified low-level events, e.g.…”

“…In [37], the length of the window is dynamically derived at runtime based on the occurrence of specified low-level events, e.g. the change of sensor state.…”