Advanced Smartphone Pedometer using Accelerometer Data
Vivek S1, Mahalingam P R2
U.G Student, Computer Science and Engineering, Muthoot Institute of Technology and Science, Kochi, India1
Assistant Professor, Computer Science and Engineering, Muthoot Institute of Technology and Science, Kochi, India2

Abstract: Smartphones are literally a part of our daily life now a days, we carry it all times along us. The purpose of this paper is to describe the design of an algorithm which can basically make your smartphone act as a pedometer by utilizing the data coming from accelerometer. As a part of this we have build an Android application featuring the algorithm. The application, while running in background collects the 3-axis accelerometer data, finds the total acceleration occurring using a set of mathematical functions and then provides user with various insights from collected data. While most smartphone pedometer algorithms count “false steps”, caused by random movements of the device, our algorithm aims to resolve this problem. In this paper we discuss the methods for counting steps, how “false step” counting is prevented and details of the Android application we built.

Keywords: Smartphone, Pedometer, Accelerometer, Android Application

I. Introduction
No computing device has been adopted like a smartphone. They are ubiquitous. One of the distinct features of a smartphone compared to any other computing platform is that, smartphones comes with a host of sensors like accelerometer, proximity sensor, gyroscope, hall effect sensor etc. While not all of these sensors are found in smartphone owing to device fragmentation issues, sensors like accelerometer sensors are found on every smartphone.
And this creates another question, are we really utilizing these sensors efficiently? For example, we do make use of accelerometer sensor for gaming purposes. But it's a motion sensing sensor, and can be used for other practical applications also. Can we apply it in to health care? This project makes use of accelerometer sensor, that is a 3-axis accelerometer sensor found commonly in smartphones to detect physical activities of users such as walking, running[1] etc and make out valuable insights to these findings such as step-counting. As a result, we have an Android application capable of doing this. We choose Android as our platform owing to it's popularity, openness etc.
The purpose of this paper is to describe an algorithm we developed which can track user physical activity and number of steps taken, by analyzing accelerometer data. In order to implement the algorithm we collected accelerometer sensor data[5] from different users as they performed activities like walking, running, standing still while wearing the phone in their pants pocket. We optimized the data and then classified the physical activities based on amplitude of frequencies present in accelerometer data.
This algorithm is implemented and tested as an Android application, which collects accelerometer data while running in background[4]. Here user’s physical activity is being recognized by comparing the amplitude of frequencies present in accelerometer data to standard amplitude values which we find out by collecting data from various users. In order to avoid counting “false-steps”, algorithm contentiously checks what movement to be count as a “step” and what’s not.
The hardware used for this project is a smartphone running on Android 5.0.2. It has an inbuilt 3-axis accelerometer which we utilized. The software required for developing the application was Android Integrated Development Environment(IDE) which comes with Android Software Development Kit(SDK). It provides all software needed to develop Android applications.

II. Description
The algorithm we used is defined here. A flowchart of the algorithm is given along side with it. This algorithm was developed after closely observing the data we collected as a part of this project.
A. Accelerometer Data The accelerometer data is collected from smartphone. For this purpose we used a Motorola moto-g[7] phone which has a built in 3-axis accelerometer. The data is collected using SensorEventListener functionality of Android SDK.This data is collected with sensor delay set to SENSOR_DELAY_FASTEST within the android application so that readings are taken every 10ms. For application to run in background data collection is done as a service. In order to have data collection going on uninterruptedly, service is set as STICKY_SERVICE so that it will restart itself if terminated other than by the user.
The accelerometer data will be the amount of acceleration happening to smartphone about x, y, z axes and will be as 3 separate values.

B. Analyze data
Since it is a 3-axis accelerometer, the data collected is along 3-axises. That is the acceleration in 3-axises. We find the total acceleration out this using equation.

Z=√(X2+Y2+Z2)

where Z is the total acceleration. This accelerometer data would look like following when plotted.
[pic]
we can see from the chart the data seems to produce similar looking vibrations. In order to get a more useful idea of this data, we find the amplitude of these vibrations. This amplitude analyzed to recognize what activity the user is carrying on with[2]. This is possible since amplitude may vary greatly according to different activities.

C. Activity Recognition During this stage the analysis of amplitude of vibration we received in previous stage take place. The amplitude is checked to see which activity it belongs. When no activity is taking place, amplitude will be small. While user is walking amplitude begins to grow. It will be in a moderate level here. When he is running amplitude will have higher values. During the initial stages of building this algorithm, we collected data from various people having different body dimensions and found out how amplitude is varied according to activity. The results are that

• The amplitude may vary between 0 to 4 if the user is not moving at all. • The amplitude may vary between 6 to 18 if the user is walking. • The amplitude may vary between 25 to 45 if the user is running.

Now we can determine which activity is user currently involved in. Then it is checked to if the activity is real or false one. That is user is actually carrying out this activity or whether this activity is detected due to system detecting some false readings. This checked and if the activity is recognized due to false readings, we go back to 1st stage to get new accelerometer data and ignore the current detection.

D. Update values in App Once the system detects true activities, the application UI is updated to bring user updated step count and current activity being carried out. However if the application is running in background UI can not be updated and hence corresponding values are updated.

III. Avoiding False-Steps This kind of activity recognition has a serious drawback. In fact all smartphone activity recognition systems has this same serious drawback. Random movements of the user which does not belong to any specific activity may be detected by the system and classified as one of the activity. In case of smartphone pedometer this may be counting of “false-steps”. “false-steps” are steps that are not actually taken by the user but are counted by the system. They arise mainly due to hand movement of user holding the smartphone. This can be easily identified, as they may not have the natural vibration pattern like a real activity would do. The amplitude of this vibration pattern will vary drastically in each second. Successive amplitude values will not be in same range[8].

Any reading cannot be detected to be false or not, if only one value is considered at a time. So we compare current detection with previous detection to make sure that user is naturally doing the activity. That is if user is really walking then successive amplitude of vibrations will be in a similar range. The same goes while user is running or standing still. So any drastic back and forth change can be regarded as an anomaly and can be counted false. So for such readings the step-count is not updated and readings are simply ignored.

IV. Android Application The Android application which implement above said algorithm was built. It is compatible from Android version 4.0.3 onwards although the preferred Android version is Android 5.0.2 .

By this almost 85% Android devices are able to run this application. It has a simple user interface. The start screen tells users what the application can do.

Once started, the application displays how many steps are taken and what activity is user currently engaged(Still / Running / Walking)[6]. This may be improved further using machine learning[3].

V. Conclusion
Here an algorithm for detecting physical activities of user was built. It currently works as a pedometer. The algorithm is smart enough to detect and avoid “false-step” counts that can occur. The algorithm was implemented using an Android application that can be run on most of the android smartphones.
It has an accuracy of about 90% after taking into consideration the possibility of bugs in the implementation. This may be improved by using machine learning and analyzing usage patterns to arrive at definite conclusions.

References
1] Tudor-Locke, Catrine, et al. "Comparison of pedometer and accelerometer measures of free-living physical activity." Medicine and science in sports and exercise 34.12 (2002): 2045-2051.
2] Schneider, Patrick L., Scott E. Crouter, and DAVID R. Bassett. "Pedometer measures of free-living physical activity: comparison of 13 models." Medicine and Science in Sports and Exercise 36.2 (2004): 331-335.
3] Lee, Young-Seol, and Sung-Bae Cho. "Activity recognition using hierarchical hidden markov models on a smartphone with 3D accelerometer." Hybrid Artificial Intelligent Systems. Springer Berlin Heidelberg, 2011. 460-467.
4] Shyi-Shiou, Wu, and Wu Hsin-Yi. "The design of an intelligent pedometer using Android." Innovations in Bio-inspired Computing and Applications (IBICA), 2011 Second International Conference on. IEEE, 2011.
5] Shin, Jaewan, et al. "Human movement detection algorithm using 3-axis accelerometer sensor based on low-power management scheme for mobile health care system." Advances in Grid and Pervasive Computing. Springer Berlin Heidelberg, 2010. 81-90.
6] Oshin, Thomas Olutoyin, and Stefan Poslad. "ERSP: An Energy-Efficient Real-Time Smartphone Pedometer." Systems, Man, and Cybernetics (SMC), 2013 IEEE International Conference on. IEEE, 2013.
7] Park, Tae Hyang, and Hee-Cheol Kim. "From Mechanical Pedometer to Digital Pedometer: A Usability Study on a Walking Promotion System."
8] Zeng, Qingchi, et al. "A Simple Step Counting Method Based on Smartphone Acceleration Sensors." 한국통신학회 학술대회논문집 (2014): 87-88.