Multi-resident activity recognition using label combination approach in smart home environment

Activity recognition in smart home environment is becoming challenging when it is involving more than one resident living in the same space. It is not merely recognizing the activity performed nevertheless to track and identify the performer of specific activity also need to address in order to provide the great autonomous for ambient intelligence system (Ami). It is a challenging task due to diversity and complexity of sensor fusion that only using the binary data from single type technology of ambient sensors. Strong approach is needed to identify types of activities performed at the same time to track which resident are performing that particular activity. Previously, researchers build the multi-resident activity model regardless the performer, thus the data association also fails to tackle the problem applicably. This research presents the multi-label classification approach to recognize the activity at the same is able to track the resident in multi-resident in a smart home setting. It has been tested on the real smart home datasets using Label Combination method of multi-label classification technique using random forest as its base classifier. The Hamming score, accuracy and exact match are selected as evaluation metrics to measure the proposed solution.

I. Introduction

Sensors and actuators are facing technology advancement in pervasive sensing infrastructure in term of low cost, low power and affordable but high performance sensors. Difference types of sensors have different functions and applications. There are wearable sensors, ambient sensors and vision-based sensor such as video camera and camera that use for human activity recognition. The selection of the sensor type plays an important role so that the implementation is accepted and practical by the resident. Nowadays, most of elderly adult prefer to stay at own home rather than rehabilitation center. They are seems independent to do their activities of daily living (ADL). However, there is a need to monitor their activities from distance by their children, caregiver or nurse. Therefore, the selection of sensor is significant to consider because elderly adult are naturally weak, senile and sometimes reluctant to wear if they are given wearable sensor to carry around every day. This indication of strongly emphasize that the ambient sensor is the most appropriate sensor selection to install within their smart home environment. Fig. 1.

Sensor and contact layer to infer resident activity.