Table 5.

Summary of research works that use sound sensor technology.

Research Authors (Year)	Purpose	Characteristics	Outcomes
Fleury et al. [102] (2008)	Walking, bending, and sitting recognition	The work proposed stimulating activities in a laboratory setting. The case study considered one day of monitoring.	100% accuracy.
Khan et. al [103] (2015)	Fall detection	The proposed research work developed a fall detection system based on acoustic signals collected from elderly people while performing their normal activities. The authors constructed a data description model using source separation technique, Mel-frequency cepstral coefficient, and support vector machine to detect falls. The dataset used in the work consisted of 30 fall activities and 120 non-fall activities.	100% accuracy
Li et al. [104] (2010)	Fall detection	The proposed work presented an eight-microphone circular array for person tracking and fall detection. For the sound classification, the authors applied Mel-frequency cepstral coefficients. Main design features of the array were obtained by utilizing a simulation toolbox in MATLAB.	100% accuracy
Li et al. [105] (2011)	Fall detection and localization	The authors proposed an approach for improving the accuracy of acoustic fall detection based on sliding window position and duration in data. The authors found that by positioning the window at the starting position of the signal, the highest sound source localization performance was achieved. This work applied the Hilbert transform by using a finite impulse response filter on the signals.	100% accuracy.
Popescu et al. [106] (2008)	Fall detection	Five different types of falls were targeted for experiments. A nurse was assigned to direct the subjects during the recording sessions of falls. The experimental dataset consisted of six different sessions with 23 falls in total.	100% accuracy.
Popescu & Mahnot [107] (2009)	Fall detection	The proposed work investigated a one-class classifier that required only examples from one class (i.e., fall sounds) for training. Then, fall detection was carried out based on that training.	100% accuracy.
Vacher et al. [108] (2011)	Recognition of activities of daily living	The work proposed Gaussian mixture models and support vector machines for daily activity recognition. The system also tried to recognize significant events rather than daily activities.	92% accuracy.
Zhuang et al. [109] (2009)	Fall detection	The author presented a fall detection system that used only the audio signal of the microphone. The system modeled each fall segment using a Gaussian mixture model super vector. A support vector machine was combined with the model supervisors to classify audio segments into falls.	64% accuracy.