- This paper presents Corredor, a human-centricsensing system that encourage people's physical activity. The main objective of Corredor is to help people, that suffer obesity, during their workout as part of their treatment. Corredor uses phone's embedded sensors along with machine learning algorithms to recognize human activities such as running, walking and standing. Corrredor runs enterally in the user's phone and does not require any external server processing. In addition, Corredor displays on the screen the followed route by the user, indicating the segments where the user was running, walking or standing. The system computes a set of 64 features from realtime accelerometer data using a 5 seconds sliding window with 50% of overlapping. The computed features are used to train a C4.5 decision tree which in turns is used to recognize workout activities. After system evaluation, our results show that Corredor achieves up to 93.7% overall accuracy. Finally, the application saves the historical data and is able to show them using Google Maps.
Advancements in pervasive computing are rapidly changing
preventative healthcare. Under the status quo, the average
healthy individual visits the doctor rarely, perhaps just once a
year. The doctor assesses the patient and then may prescribe
medications and recommend behavior changes (reduce fat
consumption, exercise more, etc.). One year later, the patient
returns and this process is repeated. In the emerging new
model of health care, the patient carries sensors that monitor
health in real-time, as the patient goes about normal daily life
[
Physical activity is considered a preventive mechanism to
avoid and control problems such as obesity and psychological
stress. Both are well know issues in public health. Obesity is a
leading cause of death worldwide, with increasing prevalence
in adults and children. Obesity-related conditions include heart
disease, stroke, type 2 diabetes and certain types of cancer.
Medical costs associated with obesity were estimated at $147
billion; the medical costs for people who are obese were
$1,429 higher than those of normal weight [
Taking these facts into consideration, in this paper, we present Corredor, human-centric sensing system for activity tracking and recognition with application in preventive health. Physical activity is considered a preventive mechanism to avoid and control problems such as obesity and psychological stress. Both are well know issues in public health. The main idea is to employ persuasive and behavioral techniques to keep the patient engaged and motivated to meet health goals.
Corredor is a mechanism that allows people to track their
workout progress using smart phones which has potential
application in mHealth. Given the fact that people use their
phones on a daily basis and carry them almost every place,
this is an illustrious technology that could potentially help
solve this health epidemic. However, the sensor raw data are
not sufficient in order to identify people’s behavior. One of the
key challenges in creating useful and robust ubiquitous
applications is context detection from noisy and often ambiguous
sensor data [
Our application allows users to track their running, walking, or standing activities. The system has two modules, the activity recognition module, and the visualization module. The first recognizes, and reports to the user the performed activities and their time duration; while the second module uses the phones GPS and Wifi sensor to collect outdoor and indoor location data, and allows users to track the followed route during her workout showing the segments running, walking and standing. This feature allows users to plan their route in terms of goals during their workout.
The rest of the paper presents the related work to this project followed by the system description, the experimental settings and results. Finally, the conclusions are presented along with some considerations for future research in this area.
Copyright © 2015 for the individual papers by the papers’ authors. Copying permitted for private and academic purposes. This volume is published and
copyrighted by its editors. Latin American Workshop On Communications' 2015 Arequipa, Peru Published on CEUR-WS: http://ceur-ws.org/Vol-1538/
and their surroundings. In addition, there are available external three two modules: collector module and the classification
sensors equipped with communication capabilities which allow module. The collector application collect ground true data,
their integration with other mobile devices within Personal which is used by the tester module to build the classier that will
Area Networks (PANs) or Body Area Networks (BANs) [
On the other hand, human activity recognition has became their interrelationships. The following are the main elements
a useful tool for military, security, and, especially, for medical of the Corredor.
applications [
Although several applications have been proposed for
human activity recognition using smart phone, many of them
require additional devices such as external straps that the
patient must wear in order to sense data. This is the case
of Centinela which requires the BioHarnessT M BT chest
sensor strap manufactured by Zephyr [
We design an android application that allows the users to track their running, walking, or standing activities. Users can chose whether to manually input data or to use automatic recognition module. These tasks can be used all day long automatically or manually activated, see Figure 1.
The system is organized in two main modules, the activity recognition module, and the visualization module. The Corredor’s activity recognition module is in turns subdivided in the
We created an Android application for data collection, the application uses the phone’s accelerometer sensor for activity recognition, and GPS for visualization. We collect the three values associated with accelerometer data, namely the axes x,y, and z at a sampling rate of 50Hz. On average, sensor values were received every 5-10 ms. The data ground true collection was performed by a single individual for running, walking, and still. For running and walking, the phone was held in the hand in various positions to simulate possible reallife scenarios. For sitting still, the phone was in the pocket and recorded data during normal desk work. Figure 3
We compute a set of 64 features, 63 in the frequency domain, and one in the time domain. Every time that we obtain a new (x; y; z) acceleration sample we compute its magnitude m using Equation1 m = px2 + y2 + z2 (1)
We buffer up 64 consecutive magnitudes, namely, fm0; : : : ; m64g and compute the Fast Fourier Transform,(FFT) of each element in order to form a new frequency vector with elements ff0; : : : ; f63g. Finally, the last feature corresponds to maxa = maxfm0; : : : ; m64g, forming the feature vector ff0; : : : ; f63; maxag.
The data was divided into five-second time windows. We implemented the concept of sliding time windows, which overlapped by 50% as shown in Figure 4. Sliding time windows are widely known to reduce classification error during activity transition.
Using the collection mechanism described ins section III-A we build a ground true with label features of three activities as show Figure 5.
We download the ground true data from the phone and use Weka to build a used the ground true to generate a J48 prune decision three as shown in Figure6
The resulting classifier, namely the jar file is include as a subroutine of the phone application and used along with the FFT subroutine for classification in the production stage as showed in Figure 7.
The accuracy of the classifier was evaluated using a customized form of stratified ten-fold cross validation. Ten-fold cross validation randomly splits the testing set into ten equally sized subsets. The folds are stratified, which means each fold contains a proportional amount of each class. For each fold, we train on the other nine folds and test on the current fold, and average together each folds classification accuracy for a total predicted accuracy. Table I presents the confusion matrix, here the elements of main diagonal are significatively bigger than the elements out of diagonal showing a low level of false positives and true negatives. Table II shows the detail accuracy per class, and its last line presents the weight average over the three activist. Finally, Table III presents a shows the number of correctly and incorrectly classified instances as well as the mean and absolute classification errors. of the computed statistical error estimation.
In this work, we explore a preliminary approach to save energy based on a modification of the popular C4.5 algorithm. The main idea behind this modification is to take into account not only information gain as a criteria for branch partition but also energy consumption. The following section sketch the main components of our approach.
The Power-Aware Decision Tree algorithm (PAT) considers
the sensors’ power consumption along with feature’s
information gain in order to increase the accuracy of the activity
recognition process as well as the power efficiency. PAT
is based on the popular C4.5 algorithm developed by Ross
Quinlan, which greedily chooses splits on attributes to build a
decision tree by maximizing information gain [
C4.5 uses the concept of information entropy to calculate
the level of uncertainty of an attribute split and compare it
with the information entropy without the split. The
KullbackLeibler (KL) divergence (also known as information gain) is
the difference between those two information measures, and is
used as the criterion to generate the splits while the decision
tree is being built. The KL divergence is a way of comparing
two probability distributions, and is defined as follows [
butions q(x) and p(x):
KLqjp hlog q(x) log p(x)iq(x) 0
We introduce a new criterion for split selection that takes into account not only the KL divergence, but also the knowledge of sensor power efficiencies. The main idea is to create a tree that favors a combination of the most power efficient and the most informative attributes. Table IV shows the weights assigned to each of the sensors that were used, with 1 being the least power efficient and 10 being the most power efficient. In actual applications, these weights would correspond to the relative power efficiencies of the sensors.
We introduce a new criterion for split selection that takes into account not only the KL divergence, but also the knowledge of sensor power efficiencies. The main idea is to create a tree that favors a combination of the most power efficient and the most informative attributes. Table IV shows the weights assigned to each of the sensors that were used, with 1 being the least power efficient and 10 being the most power efficient. In actual applications, these weights would correspond to the relative power efficiencies of the sensors. It is important to note that in our experiments we did not assign realistic weights to the sensors...we assigned these weights so that we could test the behavior of the algorithm. In actual applications, these weights would correspond to the relative power efficiencies of the sensors.
Like C4.5, PAT chooses splits by finding the attribute that will maximize the split criteria. The split criteria is a linear combination of the Kullback-Leibler divergence and the power efficiency of the attribute’s associate sensor. We control the relative weights of the KL divergence and the power efficiency with a parameter . This new split criteria S is defined as follows:
This paper presents Corredor, a human-centric sensing platform for human activity recognition based upon human acceleration data. An extensive evaluation was performed for a set of 64 features, a J48 decision tree, eight classification, and 5 seconds sliding window with a 50% of overlap . Overall, the mean accuracy achieved was 93.2%. This result supports the hypothesis that a energy efficient system based on only acceleration data are enough to reach high labels of activity recognition accuracy.