# Fitness Activity Recognition on Smartphones Using Doppler Measurements

^{1}

^{2}

^{*}

^{†}

## Abstract

**:**

## 1. Introduction

## 2. Physical Principles and Preprocessing Algorithm

## 3. Hardware Limitations and Placement

## 4. Classification Methods and Evaluation

#### 4.1. Classical Machine Learning Classification Schemes

**Naive Bayes (NB) Classifier**is built on the Bayes theorem. It is a supervised learning algorithm. The assumption here is that every feature pair is independent. It is the simplest form of Bayesian network. Bayes theorem provides a way to calculate the posterior probability $P\left(c\right|x)$ from the probabilities $P\left(c\right)$, $P\left(x\right)$ and $P\left(x\right|c)$ of the sample distribution by applying Equation (2).

**Support vector machines (SVM)**are known as a large margin linear classifier. A hyperplane separates the classes in the SVM algorithm. The SVM works by finding the maximized distance between the hyperplane boundaries. The distance between the boundaries is called the margin. However, unlike logistic regression for binary classification, the support vector machine does not provide probabilities, but only outputs a class identity. SVM can also classify non-linear data. The process of doing this is through a process called kernel trick. Here, the SVM transforms the non-linear data from a lower dimension to a higher dimension and tries to linearly separate the data there. SVM is, in general, a binary classification scheme, but it can be extended to a multi-classes algorithm by using the one-versus-rest or the one-versus-one method. The theory of SVM can be found in various textbooks [24].

**Random Forests**work by training many decision trees on random subsets of the features, then averaging out their predictions. Building a model on top of many other models is also called Ensemble Learning, and it is often a great way to reduce the problem of over-fitting. The performance of Random forest is always better than the individual decision trees they rely on. Since the underlying single classifiers are independent from each other, they can be trained individually and hence fasten the learning speed. This kind of Ensemble Learning is also called bagging.

**AdaBoost:**One way for a new predictor to improve its predecessor is to pay a bit more attention to the training instances that the predecessor under-fitted or wrongly classified. This results in new predictors focusing more and more on the difficult cases. It keeps going until the number of the predefined models is arrived and then the perfect linear combination is constructed to classify the underlying problem by using Equation (3).

#### 4.2. Evaluation and Comparison

#### 4.3. Activity Recognition Based on Convolutional Neural Networks

## 5. Conclusion and Outlook

## Author Contributions

## Conflicts of Interest

## References

- Anderson, F.; Grossman, T.; Matejka, J.; Fitzmaurice, G. YouMove: Enhancing Movement Training with an Augmented Reality Mirror. In Proceedings of the 26th Annual ACM Symposium on User Interface Software and Technology, St. Andrews, UK, 8–11 October 2013; pp. 311–320. [Google Scholar]
- Velloso, E.; Bulling, A.; Gellersen, H. MotionMA: Motion Modelling and Analysis by Demonstration. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, Paris, France, 27 April–2 May 2013; pp. 1309–1318. [Google Scholar]
- Kirchbuchner, F.; Grosse-Puppendahl, T.; Hastall, M.R.; Distler, M.; Kuijper, A. Ambient Intelligence from Senior Citizens’ Perspectives: Understanding Privacy Concerns, Technology Acceptance, and Expectations. In Ambient Intelligence; Springer: Berlin, Germany, 2015; pp. 48–59. [Google Scholar]
- Ding, H.; Shangguan, L.; Yang, Z.; Han, J.; Zhou, Z.; Yang, P.; Xi, W.; Zhao, J. FEMO: A Platform for Free-weight Exercise Monitoring with RFIDs. In Proceedings of the 13th ACM Conference on Embedded Networked Sensor Systems, Seoul, Korea, 1–4 November 2015; pp. 141–154. [Google Scholar]
- Mitchell, E.; Monaghan, D.; O’Connor, N.E. Classification of Sporting Activities Using Smartphone Accelerometers. Sensors
**2013**, 13, 5317–5337. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Schmidt, A. Implicit human computer interaction through context. Pers. Technol.
**2000**, 4, 191–199. [Google Scholar] [CrossRef] - Lu, H.; Pan, W.; Lane, N.D.; Choudhury, T.; Campbell, A.T. SoundSense: Scalable Sound Sensing for People-centric Applications on Mobile Phones. In Proceedings of the 7th International Conference on Mobile Systems, Applications, and Services, Kraków, Poland, 22–25 June 2009; ACM: New York, NY, USA, 2009; pp. 165–178. [Google Scholar]
- Schweizer, I.; Bärtl, R.; Schulz, A.; Probst, F.; Mühlhäuser, M. NoiseMap-Real-time participatory noise maps. In Proceedings of the Second International Workshop on Sensing Applications on Mobile Phones, Seattle, WA, USA, 1–4 November 2011. [Google Scholar]
- Popescu, M.; Li, Y.; Skubic, M.; Rantz, M. An acoustic fall detector system that uses sound height information to reduce the false alarm rate. In Proceedings of the 30th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Vancouver, BC, Canada, 20–24 August 2008; pp. 4628–4631. [Google Scholar]
- Fu, B.; Karolus, J.; Grosse-Puppendahl, T.; Herrmann, J.; Kuijper, A. Opportunities for Activity Recognition using Ultrasound Doppler Sensing on Unmodified Mobile Phones. In Proceedings of the 2nd international Workshop on Sensor-based Activity Recognition and Interaction, Rostock, Germany, 25–26 June 2015. [Google Scholar]
- Gupta, S.; Morris, D.; Patel, S.; Tan, D. SoundWave: Using the Doppler Effect to Sense Gestures. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, Austin, Texas, USA, 5–10 May 2012; pp. 1911–1914. [Google Scholar]
- Aumi, M.T.I.; Gupta, S.; Goel, M.; Larson, E.; Patel, S. DopLink: Using the Doppler Effect for Multi-device Interaction. In Proceedings of the 2013 ACM International Joint Conference on Pervasive and Ubiquitous Computing, Zurich, Switzerland, 8–12 September 2013; pp. 583–586. [Google Scholar]
- Sun, Z.; Purohit, A.; Bose, R.; Zhang, P. Spartacus: Spatially-aware Interaction for Mobile Devices Through Energy-efficient Audio Sensing. In Proceedings of the 11th Annual International Conference on Mobile Systems, Applications, and Services, Taipei, Taiwan, 25–28 June 2013; pp. 263–276. [Google Scholar]
- Yang, Q.; Tang, H.; Zhao, X.; Li, Y.; Zhang, S. Dolphin: Ultrasonic-Based Gesture Recognition on Smartphone Platform. In Proceedings of the 2014 IEEE 17th International Conference on Computational Science and Engineering (CSE), Chengdu, China, 19–21 December 2014; pp. 1461–1468. [Google Scholar]
- Ruan, W.; Sheng, Q.Z.; Yang, L.; Gu, T.; Xu, P.; Shangguan, L. AudioGest: Enabling Fine-grained Hand Gesture Detection by Decoding Echo Signal. In Proceedings of the 2016 ACM International Joint Conference on Pervasive and Ubiquitous Computing, Heidelberg, Germany, 12–16 September 2016; ACM: New York, NY, USA, 2016; pp. 474–485. [Google Scholar]
- Nandakumar, R.; Gollakota, S.; Watson, N. Contactless Sleep Apnea Detection on Smartphones. In Proceedings of the 13th Annual International Conference on Mobile Systems, Applications, and Services, Florence, Italy, 18–22 May 2015; ACM: New York, NY, USA, 2015; pp. 45–57. [Google Scholar]
- Nandakumar, R.; Iyer, V.; Tan, D.; Gollakota, S. FingerIO: Using Sonar for Fine-Grained Finger Tracking. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, San Jose, CA, USA, 7–12 May 2016. [Google Scholar]
- Xi, W.; Huang, D.; Zhao, K.; Yan, Y.; Cai, Y.; Ma, R.; Chen, D. Device-free Human Activity Recognition Using CSI. In Proceedings of the 1st Workshop on Context Sensing and Activity Recognition, Seoul, Korea, 1 November 2015; pp. 31–36. [Google Scholar]
- Fu, B.; Gangatharan, D.V.; Kuijper, A.; Kirchbuchner, F.; Braun, A. Exercise Monitoring On Consumer smartphones Using Ultrasonic Sensing. In Proceedings of the 4th International Workshop on Sensor-based Activity Recognition and Interaction, Rostock, Germany, 21–22 September 2017; ACM: New York, NY, USA, 2017. [Google Scholar]
- Shan, X.J.; Yin, J.Y.; Yu, D.L.; Li, C.F.; Zhao, J.J.; Zhang, G.F. Analysis of artificial corner reflector’s radar cross section: A physical optics perspective. Arabian J. Geosci.
**2013**, 6, 2755–2765. [Google Scholar] [CrossRef] - Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.; Thirion, B.; Grisel, O.; Blondel, M.; Prettenhofer, P.; Weiss, R.; Dubourg, V.; et al. Scikit-learn: Machine Learning in Python. J. Mach. Learn. Res.
**2011**, 12, 2825–2830. [Google Scholar] - Smith, G.E.; Woodbridge, K.; Baker, C.J. Naíve Bayesian radar micro-doppler recognition. In Proceedings of the 2008 International Conference on Radar, Adelaide, SA, Australia, 2–5 September 2008; pp. 111–116. [Google Scholar]
- Ritchie, M.; Fioranelli, F.; Borrion, H.; Griffiths, H. Multistatic micro-doppler radar feature extraction for classification of unloaded/loaded micro-drones. IET Radar Sonar Navig.
**2017**, 11, 116–124. [Google Scholar] [CrossRef] - Steinwart, I.; Christmann, A. Support Vector Machines; Springer Publishing Company, Incorporated: Berlin, Germany, 2008. [Google Scholar]
- Ronao, C.A.; Cho, S.B. Human Activity Recognition with Smartphone Sensors Using Deep Learning Neural Networks. Expert Syst. Appl.
**2016**, 59, 235–244. [Google Scholar] [CrossRef] - Nair, V.; Hinton, G.E. Rectified Linear Units Improve Restricted Boltzmann Machines. In Proceedings of the 27th International Conference on International Conference on Machine Learning, Haifa, Israel, 21–24 June 2010; pp. 807–814. [Google Scholar]
- Tieleman, T.; Hinton, G. Lecture 6.5—RmsProp: Divide the gradient by a running average of its recent magnitude. COURSERA Neural Netw. Mach. Learn.
**2012**, 4, 26–31. [Google Scholar]

**Figure 1.**The three different sports activities monitored: bicycle exercise (

**left**), squat exercise (

**middle**) and toe touch exercise (

**right**). The placement of the smartphone is close to the wall due to a stronger back-reflection.

**Figure 2.**Bottom

**left**shows the placement of the device on the corner of the table and its spectrogram is shown on the top

**left**side. Bottom

**right**: The device was placed in the middle of the table and its corresponding spectrogram is shown on the top

**right**side. The back reflection of the corner placement is stronger due to the multi-path reflection.

**Figure 3.**The confusion matrix for different classifiers are depicted. Top left (

**1**) shows the confusion matrix for the Naive Bayes Classifier. Top right (

**2**) shows the confusion matrix for the Random Forest classifier. Bottom left (

**3**) shows the confusion matrix for the Support Vector Machine classifier. Bottom right (

**4**) shows the confusion matrix for the AdaBoost classifier. As can be clearly seen from the results, AdaBoost classifier performs better than Naive Bayes and Random Forest. Although with higher computational load, it has a worse accuracy than support vector machine. The support vector machine shows the best recognition results with fewer misclassification rate.

**Figure 4.**The ROC curve for different classifiers are depicted. Top left (

**1**) shows the ROC curve for Naive Bayes Classifier. Top right (

**2**) shows the ROC curve for the Random Forest classifier. Bottom left (

**3**) shows the ROC curve for the Support Vector Machine classifier. Bottom right (

**4**) shows the ROC curve for the Adaboost classifier. As can be extracted from the results, the Naive Bayes classifier performs the worst, which is only slightly better than random guessing. The Support Vector machine performs the best compared to the other classifiers. For all three classes, the ROC curves behaves similarly and shows the most robust results.

**Table 1.**The time duration for the fastest and slowest speed of each sports activities performed by the test participants. The abbreviation TS stands for time segment.

Exercise | Minimum (TS) | Maximum (TS) | Minimum Duration (s) | Maximum Duration (s) |
---|---|---|---|---|

Bicycle | 13 | 31 | 0.60 | 1.44 |

Squats | 12 | 42 | 0.55 | 1.95 |

Toe touches | 11 | 25 | 0.51 | 1.16 |

Estimators | Max Features | Tree Depth |
---|---|---|

300 | sqrt | 100 |

Kernel | $\mathit{\gamma}$ | Penalty Parameter |
---|---|---|

Linear | 0.0001 | 1 |

True Label | Predicted Label | |
---|---|---|

Positive Sample | Negative Sample | |

Positive Sample | TP | FN |

Negative Sample | FP | TN |

Naive Bayes | Random Forest | Support Vector Machine | AdaBoost | |
---|---|---|---|---|

Precision | 64% | 77% | 84% | 77% |

Recall | 61% | 68% | 83% | 75% |

Accuracy | 60% | 72% | 83% | 76% |

Naive Bayes | Random Forest | Support Vector Machine | AdaBoost | |
---|---|---|---|---|

Precision | 51% | 68% | 74% | 67% |

Recall | 51% | 60% | 70% | 61% |

Accuracy | 55% | 64% | 74% | 65% |

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Fu, B.; Kirchbuchner, F.; Kuijper, A.; Braun, A.; Vaithyalingam Gangatharan, D.
Fitness Activity Recognition on Smartphones Using Doppler Measurements. *Informatics* **2018**, *5*, 24.
https://doi.org/10.3390/informatics5020024

**AMA Style**

Fu B, Kirchbuchner F, Kuijper A, Braun A, Vaithyalingam Gangatharan D.
Fitness Activity Recognition on Smartphones Using Doppler Measurements. *Informatics*. 2018; 5(2):24.
https://doi.org/10.3390/informatics5020024

**Chicago/Turabian Style**

Fu, Biying, Florian Kirchbuchner, Arjan Kuijper, Andreas Braun, and Dinesh Vaithyalingam Gangatharan.
2018. "Fitness Activity Recognition on Smartphones Using Doppler Measurements" *Informatics* 5, no. 2: 24.
https://doi.org/10.3390/informatics5020024