A lightweight and aggregated system for indoor/outdoor detection using smart devices
Graphical abstract
Introduction
Recent years have witnessed a prosperous development of location-based service, mobile sensing and context-aware detection. The boom has benefited from the changes in the concept of life and a large number of smart devices. Most of the mobile market growth can be attributed to the increasing popularity of smartphones. Up to now, the number of smartphone users worldwide has reached 2.1 billion, and this number is expected to reach 2.6 billion by 2019. Besides smartphones, wearable devices, such as smart glasses, smart watches, hearables, fitness and health trackers or even smart jewelry and smart fashion, also come into sight frequently. 2014 was hailed by many tech publications and experts as the “Year of the Wearable”, and the global market for wearable technology is forecast to be around three billion US dollars by 2016 according to the source [1]. With the help of numerous smart devices, users obtain access to locating and sensing. Typical scenes may happen: during a break, an employee walked out of the office building, and went shopping nearby after having lunch; a traveler checked out from a hotel, weaved in and out through the traffic, wandered in the airport and finally got on a plane. People may shuttle between buildings, and wander from inside to outside areas. There are many applications depending on the indoor/outdoor (IO) state, for example, human localization and tracking, context recognition, transportation detection, power management, medical care [2], etc. On one hand, users may interact with household electric appliances or smart TV in indoor scenarios; on the other hand, users may require the map and navigation APPs in outdoor scenarios. Besides, the screen brightness of smartphones vary from indoor to outdoor according to the different environments’ luminance. Among all the above examples, they usually pre-define the indoor/outdoor state, which may not be a priori knowledge. If we are aware of our locations, states, and conditions, we can manage mobile sensing tasks more effectively and efficiently. Many researchers have implemented a variety of methods for localization in indoor or outdoor environments. Meanwhile, the indoor/outdoor detection remains an open research field when we are aiming for seamless positioning and context-aware sensing. Indoor/outdoor detection is set up as a bridge between indoor and outdoor localization techniques, which has much potential in both academia and industry. Moreover, the improving WiFi infrastructure and precise built-in sensors may make a great contribution to the promising indoor/outdoor detection.
There are two existing directions for indoor/outdoor detection. One is to do the detection work on a single device locally (e.g., [3], [4]). The other models the problem from a machine learning point of view and usually gets help from computing servers. In the second category, IODetector [5] and [6] are quite representative. In this paper, we design a mixed system after exploring the advantages and shortcomings of both techniques. On one hand, the WiFi-only method can meet the need of lightweight detection running on only one terminal. On the other hand, leveraging a variety of sensors and apposite machine learning methods, we can achieve a remarkable accuracy in detecting indoor/outdoor environment and serve upper-layer applications better.
We devise a indoor/outdoor detection system including two main parts: the lightweight WiFi sub-detector and the machine learning based IOS-detector. “IOS” is the abbreviation for I ndoor, O utdoor, and S emi-open areas. In our approach, we define three states: I ndoor (inside a building), O utdoor (outside a building), and S emi-open (near a building), as shown in Fig. 1. In the lightweight sub-detector, we leverage the absolute values and the relative comparisons of WiFi RSS, and infer whether a user resides indoor, outdoor or semi-open according to the classification results from a weak learner approach. As a coarse-grained detector, this part focuses on the ability of running on mobile devices. The other part of our system pursues the high accuracy of IOS detection. We constructively model the IOS transition as a state recognition problem based on semi-Markov conditional random fields (semi-CRF) [7]. The heuristic idea lies in an extra consideration of the duration of activities which corresponds to the internal feature of IOS detection task. We also tap a great potential of the interdependency among activities when detecting IOS states in the continuous sensor data stream using semi-CRF algorithm. Thus, our system covers both situations of lightweight IO detection and sophisticated IOS detection, which can meet various needs of upper-layer applications.
We implement the system on the Android platform using different smartphones and evaluate the detector on 18 traces in three different scenes in the campus and commercial plaza. The experiment results show that the WiFi sub-detector is effective in lightweight indoor/outdoor detection and the IOS detector based on the creative semi-CRF model achieves a higher precision than previous methods [5], [6].
We summarize our contribution as follows:
- –
We develop a lightweight IOS detection algorithm leveraging only WiFi sensor and an AdaBoost.MH weak learner. This sub-detector can work in a low energy consumption.
- –
We aggregate the WiFi and light sub-detectors into an IOS detector. By bringing in the semi-CRF algorithm, the detection models the duration of each state and uncovers the interdependency between states, which improve the detection accuracy in an innovational way.
The rest of this paper is organized as follows. Section 2 reviews the related literature. We introduce the system design of IOS detection in Section 3. In Section 4, we propose the lightweight WiFi sub-detector. Semi-CRF based IOS detection algorithm is detailed in Section 5. We discuss the system performance in Section 7 and conclude the work in Section 8.
Section snippets
Related work
Indoor/outdoor detection technique mixes localization and tracking methods, mobile sensing applications, and context-aware activity recognition.
There are a lot of research focusing on localization and tracking in both indoor and outdoor environments. Indoor localization usually leverages RSS from WiFi, Bluetooth, or cell tower, along with various built-in sensors in smartphones [8], [9], [10], [11], [12]. As for outdoor localization technique, GPS occupies the dominant position [13], [14].
System overview
In this section, we present the system architecture of IOS detection. In Fig. 2, there are two main parts of the IOS detection system: lightweight WiFi sub-detector and machine learning based IOS-detector.
For lightweight IOS detection, we leverage only WiFi sensor and weak learner classifier on the absolute RSS values and their relative comparisons. Sensed data from the smart device are collected in both training and testing phases. After the AdaBoost.MH machine learning, we gain several weak
WiFi-based sub-detector
We introduce the WiFi based sub-detector in this section. The detection process includes three steps: data set collection, multi-class boost training, and weak learner classification.
AdaBoost is short for Adaptive Boosting. It is a machine learning meta-algorithm used in conjunction with other types of learning methods (weak learners). The output of weak learners combined into a weighted sum represents the final output of the boosted classifier. As for our IOS detection problem, the individual
Semi-Markov conditional random fields in indoor/outdoor detection
In this section, we briefly review the theory of Conditional Random Fields and its extension semi-CRF. We also present details of our proposed design, which is applied to the IOS detection based on the semi-CRF model.
Aggregated IOS detector using semi-CRF algorithm
We have already proposed a lightweight detection algorithm in Section 4. In this section, we leverage WiFi RSS signals and light intensity to build a complex IOS detector using the appropriate semi-CRF algorithm in Section 5. The light detection technique is presented first. Then, under the semi-CRF framework, the final IOS detector is aggregated with the light sub-detector and the WiFi sub-detector.
Evaluation
We evaluate the proposed approach above, by implementing a prototype system of IOS detector on the Android platform. We first explain the experiment settings and methodology, and then we present the performance of the integrated system after the detailed evaluation of each segments.
Conclusion
In this paper, we have proposed an IOS detection method including a lightweight WiFi-based sub-detector and an aggregated detector using semi-CRF algorithm. To explore the feasibility of our approach, we implemented a prototype of IOS detector and took experiments in various scenarios. The results show that our IOS detector can achieve an around 85% accuracy for the lightweight WiFi-based sub-detector and an over 96% accuracy for the aggregated IOS detector, manifesting its promising
Shengnan Li received the B.S. degree in Software Engineering from Nankai University, Tianjin, China, in August 2011. She received a Ph.D. degree in the department of Computer Science and Technology, Tsinghua University, Beijing. Her research interests are wireless sensor network, indoor localization, mobile sensing, and so on. Now she work for National Computer Network Emergency Response Technical Team/Coordination Center of China.
References (36)
- et al.
IODetector: A generic service for indoor/outdoor detection
ACM Trans. Sens. Netw. (TOSN)
(2015) - et al.
Spatio-temporal facility utilization analysis from exhaustive WiFi monitoring
Pervasive Mob. Comput.
(2015) - et al.
An energy-efficient and robust indoor-outdoor detection method based on cell identity map
Procedia Comput. Sci.
(2015) - et al.
Sound based indoor and outdoor environment detection for seamless positioning handover
ICT Express
(2015) - Statistical data,...
- et al.
Use runtime verification to improve the quality of medical care practice
- O. Canovas, P.E. Lopez-de Teruel, A. Ruiz, WiFiBoost: a terminal-based method for detection of indoor/outdoor places,...
- H. Cho, J. Song, H. Park, C. Hwang, Deterministic indoor detection from dispersions of GPS satellites on the celestial...
- et al.
A semi-supervised learning approach for robust indoor-outdoor detection with smartphones
- et al.
Semi-markov conditional random fields for information extraction
Smartphones based crowdsourcing for indoor localization
IEEE Trans. Mob. Comput.
Landmarc: indoor location sensing using active RFID
Wirel. Netw.
No need to war-drive: unsupervised indoor localization
Zee: zero-effort crowdsourcing for indoor localization
Trail: Pinpoint trajectory for indoor localization
Int. J. Distrib. Sens. Netw.
Entracked: energy-efficient robust position tracking for mobile devices
Human mobility enhances global positioning accuracy for mobile phone localization
IEEE Trans. Parallel Distrib. Syst.
Senseme: a system for continuous, on-device, and multi-dimensional context and activity recognition
Cited by (17)
Recurrent neural network based scenario recognition with Multi-constellation GNSS measurements on a smartphone
2020, Measurement: Journal of the International Measurement ConfederationCitation Excerpt :Wang et al. applied a machine learning algorithm to classify scenarios into four different types (deep indoors, light indoors, semi-outdoors and open outdoors) based on cellular signals [24]. Li et al. designed an aggregated indoor/outdoor/semi-outdoor detector based on Wi-Fi and light sensor, and achieved over 96% detection accuracy through a semi-Markov conditional random fields (CRF) algorithm [25]. Ali et al. invented a realistic and ubiquitous system (SenseIO) which can provide a fine-grained scenario detection [26].
A hybrid indoor/outdoor detection approach for smartphone-based seamless positioning
2022, Journal of NavigationCombining Context Connectivity and Behavior Association to Develop an Indoor/Outdoor Context Detection Model with Smartphone Multisensor Fusion
2024, IEEE Internet of Things JournalEliciting Joint Truthful Answers and Profiles From Strategic Workers in Mobile Crowdsourcing Systems
2023, IEEE Transactions on Mobile ComputingSmartphone-Based Indoor Localization Systems: A Systematic Literature Review
2023, Electronics (Switzerland)Deep-Learning-Based Scenario Recognition With GNSS Measurements on Smartphones
2023, IEEE Sensors Journal
Shengnan Li received the B.S. degree in Software Engineering from Nankai University, Tianjin, China, in August 2011. She received a Ph.D. degree in the department of Computer Science and Technology, Tsinghua University, Beijing. Her research interests are wireless sensor network, indoor localization, mobile sensing, and so on. Now she work for National Computer Network Emergency Response Technical Team/Coordination Center of China.
Zheng Qin is a doctoral supervisor, professor, the Director of Software Engineering and Management Research Institute and Information Institute, Tsinghua University. He is the evaluation expert of National Science and Technology Award, Ministry of Education Technology Award, Major State Basic Research Development Program (973), National High Technology Research and Development Program of China (863), National Defense 10th 5-Year Plan and Ministry of Education College Undergraduate Teaching. He is also the member of Ministry of Education E-Commerce Specialty Teaching Guidance Committee. He is now the professor of Tsinghua University and Xian Jiaotong University, and the part-time professor for many other high-education organizations including Ministry of National Supervision. He is the guest researcher of Shanxi Academy and Henan Academy, editor of International Journal of Plant Engineering and Management(E) and Journal of E-Business.
Houbing Song (M’12-SM’14) received the Ph.D. degree in electrical engineering from the University of Virginia, Charlottesville, VA, in August 2012. He is an assistant professor of electrical engineering and computer science in Department of Electrical, Computer, Software, and Systems Engineering in Embry-Riddle Aeronautical University. He is also the director in Security and Optimization for Networked Globe Laboratory (SONG Lab).
In August 2012, he joined the Department of Electrical and Computer Engineering, West Virginia University, Montgomery, WV, where he is currently an Assistant Professor and the Founding Director of both the Security and Optimization for Networked Globe Laboratory (SONG Lab, www.SONGLab.us), and West Virginia Center of Excellence for Cyber-Physical Systems sponsored by West Virginia Higher Education Policy Commission. In 2007 he was an Engineering Research Associate with the Texas A&M Transportation Institute. He is the editor of four books, including Smart Cities: Foundations, Principles and Applications, Hoboken, NJ: Wiley, 2017, Security and Privacy in Cyber-Physical Systems: Foundations, Principles and Applications, Chichester, UK: Wiley, 2017, Cyber-Physical Systems: Foundations, Principles and Applications, Waltham, MA: Elsevier, 2016, and Industrial Internet of Things: Cybermanufacturing Systems, Cham, Switzerland: Springer, 2016. He is the author of more than 100 articles. His research interests include cyber-physical systems, internet of things, cloud computing, big data analytics, connected vehicle, wireless communications and networking, and optical communications and networking.
Dr. Song is a member of ACM. Dr. Song was the very first recipient of the Golden Bear Scholar Award, the highest faculty research award at West Virginia University Institute of Technology (WVU Tech), in 2016.
Chengxiang Si is a senior engineer in National Computer Network Emergency Response Technical Team/Coordination Center of China. He received a Ph.D. degree from Department of Computer Science, University of Chinese Academy of Sciences in 2012.
Bo Sun is a professor-level senior engineer in National Computer Network Emergency Response Technical Team/Coordination Center of China. He received a Ph.D. degree from Department of Software, University of Chinese Academy of Sciences in 2004.