4.5 Automated Checklist Execution
4.5.1 Accelerometer Data Interpretation
In order to make the signals from accelerometer to be useful, this section looks into the interpretation of the signal from the accelerometer. In a review of the wearable accelerometer based motion detectors carried out by Yang et al. [89], the following capabilities of wearable accelerometer based system are listed but not limited to these:
• Activity recognition - Based on the signal from the accelerometer which is attached to human body, the activities that the subject is performing can be recognised. For example, Baek et al. [10] proposed an approach to recog-nise certain types of daily activities. Baek’s approach extracts the statistical features from the signal and classify those extracted feature using a multi-layer perceptron classifier. Similar approach can also be found in [67] where Ravi et al. compared many different classifiers to recognise different daily activities using mean, standard deviation, energy, and correlation as the fea-tures. The research in this domain mainly focus on the feature extraction and classification approach for the purpose of recognising a specific range of activities.
• Event detection -Detecting certain event is with paramount importance in some applications. One example will be the fall detection for the emergent medical rescue for elderly people. Subsequently, many efforts have been put into the research for more accurate event detection based on the wearable accelerometer. For example, Bourke et al. [15] proposed a threshold-based fall detection algorithm and evaluated it over different type of falls like forward falls and backward falls.
• Estimation of energy consumption - The daily energy consumption can be useful for maintain a healthy lifestyle. Such energy consumption can be estimated thorough measuring physical activities.
However, through reviewing existing works of using the accelerometer, it appears that existing approaches do not fit for the purpose of this research. Although it
Figure 4.8: Demonstration of the 3-axis accelerometer’s readings for each axis when the orientation of the accelerometer is changed (Quoted from [40]. In this ex-ample, it records the rolling movement when the subject is laying on bed. When the subject rolls between facing up, facing left, face down, and facing right, the mean value of the accelerometer’s X-axis and Z-axis readings for each facing direction is different compared to the other directions. As the subject is only rotating when laying on bed, the mean value of Y-axis reading generally remains unchanged for each directions.
is acknowledged that most existing approaches are more capable than what this research needs, it is not necessary to adopt an overly complex algorithm due to the consideration of resource consumption. Subsequently, a simplified approach based on Ravi’s and Baek’s approaches is used to interpret the data from the ac-celerometer. More specific, the mean values calculated from the accelerometer data for each axis are used to interpret the orientation status on each axis whereas the standard deviations of the data for each axis are used to describe the activ-ity intensactiv-ity on that axis. The effectiveness of each method in terms of the data interpretation will be justified below.
As demonstrated in [40], the mean value of the data from each axis can be used to represent the orientation of the mote. As shown in Figure 4.8, when a mote change its orientation, the mean values of the related axis readings from the accelerometer will change accordingly. In this example, it records the rolling movement when the subject is laying on bed. When the subject rolls between facing up, facing left, face down, and facing right, the mean value of the accelerometer’s X-axis and Z-axis readings for each facing direction is different compared to the other directions. As the subject is only rotating when laying on bed, the mean value of Y-axis reading generally remains unchanged for each directions. Such phenomenon is due to the gravity. As the gravity constantly exist and never changes on earth, such phenomenon can be used to represent the mote’s orientation. Furthermore, the calculation of the mean value does not require extensive calculation or is not heavy on memory. Subsequently, it is capable to run on a mote at run-time.
To interpret the mote’s motion intensity, the standard deviation of the data from the accelerometer is used. The standard deviation is a statistical method that mea-sures the average distance of a data sequence to its mean value. The calculation of standard deviation is shown in Equation 4.1. Figure 4.9 illustrates a period of signal from the accelerometer. This period of signal is captured when the subject is walking, running, climbing, using vacuum, using brush and performing situps, which all have different intensity compared to other activity. The intensity of the signals when each activity is performed can be interpreted by calculating the stan-dard deviation.
N : Total number of data point in data sequence x
Figure 4.9: Demonstration of the accelerometer’s reading when different activities are performed (quoted from [67]). In this example, when each activity is performed, the mean value of the accelerometer’s reading remains generally the same. How-ever, depending on the intensity of the activity, the standard deviation of the ac-celerometer’s reading for each activity is different from other.
¯
x: Mean value of the data x
i: The number of current data point