My act: an automatic detection of daily physical activity and calorie expenditure using smart phones

Figures & data

Figure 1.  The main flowchart of the binary tree algorithm that takes samples of the accelerometer and evaluates the variance to classify activity.

Figure 2.  Histogram of accelerometer variances (in log scale). The jumps are noted as the adaptive thresholds to distinguish different physical activities.

Figure 3.  Average sampling time (top) and its variance (bottom) obtained for all four different sampling periods provided by an Android operating system. The sampling rates arranged from lowest to highest: Normal, UI, Game, and Fastest, respectively. The average sampling times are not affected by different activities, but the variation of the sampling times with activities are clearly noted.

Figure 4.  Average time for 1% of battery consumption over four different sampling rates clearly shows that faster sampling rates results in lower battery times.

Figure 5.  Average time for 1% of battery consumption over four different duty cycles shows that higher duty cycles results in lower battery times. However, more variation in battery times is clearly noted in lower duty cycles.

Table I. Step counts obtained with pedometer and manual count

	Average self counts (SD)	Average pedometer count (SD)	% accuracy	t-value	p-value
Subject 1	83.9 (3.73)	82.5 (1.08)	1.70%	1.16	0.24
Subject 2	105.1 (6.14)	99.8 (5.92)	2.53%	2.61	0.009
Subject 3	96.84 (1.39)	103.67 (13.21)	6.58%	1.63	0.104

Table II. Percentage of step count error with various affecting factors compared with pedometer

	Subject 1		Subject 2		Subject 3
	Before calibration	After calibration	Before calibration	After calibration	Before calibration	After calibration
Default settings	50	21	35	16	45	21
Location
Waist	105	18	20	12	39	17
Shirt pocket	52	26	60	21	71	23
Accuracy settings
Low	83	30	26	22	92	17
High	32	14	32	18	56	12
Phones
Samsung	85	33	37	14	77	22
LG	83	29	67	16	84	34
Number of steps
5	48	23	33	12	79	43
15	21	18	18	17	95	23
30	98	16	43	13	87	19

Table

Activity class	Definitions
Sleeping	This activity is defined as when the smart phone is carried with the user while they perform very light activity. This class groups activities such as sleeping, napping, and sitting with light movement.
Sitting	This activity is defined as when the smart phone is carried with the user while they are carrying out a light movement or in a sedentary state. This class groups activities such as sitting at work, sitting while reading the newspaper or watching TV. A more sporadic movement pattern is expected in this activity than in the Sleeping activity.
Walking	This activity is defined as when the smart phone is carried with the user while the user is carrying out a moderate-tempo movement. This class groups activities such as walking, yoga and light dancing.
Running	This activity is defined as when the smart phone is carried with the user while the user is carrying out a high-tempo movement. This class groups activities such as running, walking at a fast pace, jumping, or playing some high-tempo sports such as badminton, tennis, and football.