Abstract—Obesity prevention and treatment as well as
healthy life style recommendation requires the estimation of
everyday physical activity. Monitoring posture allocations and
activities with sensor systems is an effective method to achieve
the goal. However, at present, most devices available rely on
multiple sensors distributed on the body, which might be too
obtrusive for everyday use. In this study, data was collected from
a wearable shoe sensor system (SmartShoe) and a decision tree
algorithm was applied for classification with high computational
accuracy. The dataset was collected from 9 individual subjects
performing 6 different activities—sitting, standing, walking,
cycling, and stairs ascent/descent. Statistical features were
calculated and the classification with decision tree classifier was
performed, after which, advanced boosting algorithm was
applied. The computational accuracy is as high as 98.85%
without boosting, and 98.90% after boosting. Additionally, the
simple tree structure provides a direct approach to simplify the
feature set.
I. INTRODUCTION
The World Health Organization (WHO) predicts that
overweight and obesity may soon become the most significant
cause of poor health replacing more traditional public health
concerns, such as under-nutrition and infectious diseases [1].
Obesity may have a significant effect on health, leading to
reduced life expectancy and increased health problems
including heart disease, type 2 diabetes, obstructive sleep
apnea, osteoarthritis and certain types of cancers [2]. In
addition to these health impacts, obesity may cause many
social stigmatization problems. Obesity is due to a sustained
positive energy balance and is typically coupled with low
level of physical activity [4] [5]. In other words, obesity may
be caused by excessive food energy intake and lack of
physical activity. A sedentary lifestyle plays a significant role
in obesity. The amount of work that is not physically
demanding is increasing worldwide. Moreover, there appear
to be declines in levels of physical activity in walking due to
mechanized transportation, and declines in energy
expenditure in housework due to laborsaving technology at
home. Studies show that there is an association between
television viewing time and the risk of obesity [6]. Therefore,
an accurate monitoring of physical activity directly helps in
the research of obesity. Monitoring of everyday life activities
may also provide detailed recommendations to people who are
seeking a healthy lifestyle.
Ting Zhang and Wenlong Tang are with Department of Electrical and
Computer Engineering, University of Alabama.
Edward S. Sazonov is with Department of Electrical and Computer
Engineering, University of Alabama, Tuscaloosa, AL 35487 USA (Phone:
205-348-1981; Fax: 205-348-6959; e-mail: esazonov@claws,eng.ua.edu)
For monitoring physical activities and allocations of
human beings, various devices and systems were proposed by
different research groups. For example, Bao and Intille [7]
mounted accelerometers on the wrist, upper arm, hip, ankle
and thigh, with the evaluation of a single best sensor location.
They achieved an accuracy of 84% in the activity recognition
for 20 different activities. Pirttikangas et al. [8] attached
accelerometers to left and right wrists, right thigh and a
necklace, with a recognition accuracy of 93% for 17 different
activities. However, those devices rely on sensors distributed
on the body might be too obtrusive for everyday use. To
develop the systems to be more convenient for real-life usage,
Zhang et al. [9] proposed a single tri-axial accelerometer
placed on the waist and achieved a classification accuracy of
80%. Long et al. [10] proposed a 3D-accelerometer in a smart
phone and achieved a recognition accuracy of 82.8%. A
wearable non-obtrusive device to reach high classification
accuracy in detecting posture activities still remains a desire
and challenge.
Various algorithms have been applied in physics activity
classification, such as Support Vector Machines [11] [12],
Artificial Neural Network (ANN) [13], Hidden Markov
Model (HMM) [14], Continuous Activity Recognition (CAR)
algorithm [10]. Researchers are still seeking for an optimal
solution that combines a computational effective algorithm
and an advanced sensor system.
In this study, data was acquired from a wearable shoe
sensor system (SmartShoe) developed previously by our
group [15]. After statistical feature computation from sensor
signals, decision tree algorithm with boosting [17] was used
for classification. This approach reached high classification
accuracy. The simple tree structure provided a direct approach
to simplify the feature set and this can help improving
computational efficiency.
II. METHODS
A. Wearable shoe sensors and data collection
The sensor system embedded into the shoes contains
sensors to collect plantar pressure data and heel acceleration
data. For each shoe, there are five force-sensitive registers
integrated in a flexible insole, positioned under heel, heads of
metatarsal bones, and the hallux. With this configuration,
differentiation of the most critical parts of the gait cycle, such
as heel strike, stance phase and toe-off can be performed. The
motion information is provided by a 3-D accelerometer
positioned on the back of each shoe. This wireless sensor
system is lightweight, minimally obtrusive and with advanced
power saving strategies [15].
Classification of Posture and Activities by Using Decision Trees
Ting Zhang, Wenlong Tang, Member, IEEE and Edward S. Sazonov, Senior Member, IEEE
34th Annual International Conference of the IEEE EMBS
San Diego, California USA, 28 August - 1 September, 2012
4353 978-1-4577-1787-1/12/$26.00 ©2012 IEEE