Mayur Sonawane1 Sahil Dhayalkar1, Siddesh Waje1, Soyal Markhelkar1, Akshay Wattamwar1,Prof. Seema C. Shrawne11-Department of Computer Engineering & Information Technology,Veermata Jijabai Technological Institute, Mumbai, IndiaAbstract:Human Activity Recognition is a subject of great research today and has its applications in remote healthcare, activity tracking of the elderly or the disables, calories burnt tracking etc. In our project, we have created an Android application that recognizes the daily human activities and calculate the calories burnt in real time. We first captured labeled triaxial acceleration readings for different daily human activities from the smartphone's embedded accelerometer. These readings were preprocessed using a median filter. 42 features were extracted using various methods. We then tested various machine learning algorithms along with dimensionality reduction. Finally, in our Android application, we used the machine learning algorithm and a subset of features that provided maximum accuracy and minimum model building time. This is used for real-time activity recognition and calculation of calories burnt using a formula based on Metabolic Equivalent.Keywords: accelerometer, human activity recognition, machine learning, smartphones.
TABLE I INITIAL DATASET accy -5.58328 -2.27928 -7.40287 -10.2951 -7.89129 -4.48195 -8.74362 -4.11803 -4.34787 -9.94073 -3.20823 -3.35188 -3.53384 -3.8403 -3.58173 -3.31358 -3.68707 -3.62961 -3.24654 -3.1795 -3.09331 accx: accelerometer reading across x-axis. accy: accelerometer reading across y-axis. accz: accelerometer reading across z-axis. activity: The activity performed (0 for idle, accx 13.9151 10.1993 14.05875 10.4866 12.58392 12.13381 12.00932 12.08593 12.92869 12.49773 8.973468 8.657434 9.155427 8.188169 8.552089 8.753201 8.66701 8.657434 8.724471 8.465898 8.331821 accz -3.60088 -0.61292 4.884171 4.405331 -0.67995 -2.94008 -0.96726 -2.15478 1.704672 4.309563 3.65834 3.476381 3.409343 2.920926 3.284845 3.093309 3.275268 3.102885 3.303998 4.118027 3.878607 activity 2 2 2 2 2 2 2 2 2 2 0 0 0 0 0 0 0 0 0 0 0
id: 11bc1d145b57159ac9cb27b6f2507129 - page: 3
2 for normal walking). The following graphs (Figure 1, Figure 2, Figure 3) depicts reflect the nature of readings of the accelerometer sensor across all 3 axes: X-axis: Activity (0-100 : Idle 101-200 : Slow Walking 201-300 : Normal Walking 301-400 : Fast Walking 401-500 : Jogging 501-600 : Running 601-700 : Jumping) Y-axis: Acceleration in m/s2 FIGURE 1 ACCELEROMETER READINGS ACROSS X-AXIS Idle (Standing) Slow walking Normal walking Fast walking Jogging Running Jumping Hence, by performing each activity for 3 minutes by all group members, we have a rich amount of labeled dataset. This labeled dataset will serve for training and testing of our model. A snapshot of a part of the dataset is shown in Table 1. FIGURE 2 ACCELEROMETER READINGS ACROSS Y-AXIS 3 Mean Variance Standard deviation FIGURE 3 ACCELEROMETER READINGS ACROSS Z-AXIS Energy
id: f2e38e9b0d5f656f4aadc8ad64f11f6e - page: 3
Interquartile range IQR = Q3 Q1 Kurtosis From the graphs, it can be inferred that activities in which mobile movement is low such as slow walking or idle results are the least amount of fluctuation in the readings while activities in which mobile movement is high such as jogging, running or jumping shows a high amount of fluctuation in the readings. Such fluctuations will be helpful for the machine learning algorithms to classify the readings. Noise Reduction For noise reduction, the readings were preprocessed with a median filter. A graph of some raw readings versus corresponding pre-processed readings using a median filter is depicted in Figure 4. X-axis: Activity Y-axis: Acceleration in m/s2 Raw readings. Pre-processed median filter.
id: f5d70b4e3455eb5fbaa9c221dafb0c19 - page: 4
Hence, for the 3 axes, we have 3 windows. For each window, we calculated the features using the above-mentioned methods to get a single value that against those 8 consecutive readings. From the same window, we derive new 8 values of those 8 consecutive readings using a Fast Fourier Transform. We then calculate the features using the above-mentioned methods to get a single value that against those 8 Fast Fourier transform readings. Hence, we have 42 features (21 features of normal readings of acceleration across x, y z-axis; 21 features of fast Fourier transform of the readings of acceleration across x, y z-axis). They are as follows: 1. meanaccx: Mean of 8 preprocessed acceleration values across x-axis. 2. meanaccy: Mean of 8 preprocessed acceleration
id: 86e873de46a6f01a2550afaa1150009c - page: 4