Sensing Home: A Cost-Effective Design for Smart Home via Heterogeneous Wireless Networks

Xiaohu Fan, Hao Huang, Shipeng Qi, Xincheng Luo, Jing Zeng, Qubo Xie, Changsheng Xie
2015 Sensors  
The aging population has inspired the marketing of advanced real time devices for home health care, more and more wearable devices and mobile applications, which have emerged in this field. However, to properly collect behavior information, accurately recognize human activities, and deploy the whole system in a real living environment is a challenging task. In this paper, we propose a feasible wireless-based solution to deploy a data collection scheme, activity recognition model, feedback
more » ... l and mobile integration via heterogeneous networks. We compared and found a suitable algorithm that can be run on cost-efficient embedded devices. Specifically, we use the Super Set Transformation method to map the raw data into a sparse binary matrix. Furthermore, designed front-end devices of low power consumption gather the living data of the habitant via ZigBee to reduce the burden of wiring work. Finally, we evaluated our approach and show it can achieve a theoretical time-slice accuracy of 98%. The mapping solution we propose is compatible with more wearable devices and mobile apps. Sensors 2015, 15, 30270-30292 aspects of research, including sensor technology, software and network architecture, machine learning algorithms, artificial intelligence and human-computer interaction. Activities of daily living (ADLs) [5] , such as sleeping, bathing, cooking, and so on, are good indicators of cognitive and physical capabilities [6] . While computer vision-based machine recognition methods are difficult to resolve due to the nature that the activity poses, a computer vision-based solution requires higher storage and process cost, plus the wiring and location selection rely on professional construction personnel. Simple binary sensors with advanced features of easy wiring, low cost, high robustness, low power consumption, strong expansion, high accuracy, and so on, seem to be enough to cover the home-level human activities detection [7] . As a solution, many research groups have shared their activity datasets [8] [9] [10] [11] [12] [13] [14] , and many accurate offline learning models [12, 13, 15, 16] have been put forward. However, so far there has been little work done on effective real-time classification and activity prediction solutions on embedded devices. In addition, most traditional smart home environments require complicated decoration and wiring work, so for putting up installations to deploy the sensors and controllers at the real home setting, the cost of re-decoration and human resources is huge, which hurts user satisfaction at the very beginning. Besides, the actual accuracy of activity recognition is low because the home environment settings and habits of each householder vary, so there is no general solution but to use actual activities' living data as the training set for pattern recognition. In this paper, we will propose a practical method to deploy a perceived family environment based on ZigBee, Wi-Fi and sensor technologies such as a private smart space. In this way, the traditional wiring and decoration complexity were reduced and the system is easily installed and attached to the wall, and able to work with batteries. Humidity, gas, smoke, temperature, pressure, illumination and passive infra-red (PIR) sensors are applied for designing and developing our sensing home system in this paper. We tried to monitor the behaviors of the residents to analyse the abnormal events, such as heart attacks, hypertension or falls, to inform the appropriate people to check the situation and administer first aid if necessary. A system prototype is proposed and the experimental results are discussed, including a new type of ADL records, involving time value in the computation in order to enhance the accuracy, and choose a portable algorithm for behavior recognition and feedback control policy. Thus, a cost-effective ARM-based home server capable of working with lower power consumption is devised. A mobile app for Android is provided, which integrates initial configuration, electrical appliances control, an alarm message, environmental status and optional vision-based surveillance together, and it should prove compatible with many wearable devices in the future. The main contribution of this paper is: (1) its focus on features for elderly people, comparing many famous smart home schemes and figuring out two different binary ADL datasets, and running various algorithms to find the key factors that affect the discriminant household behavior; (2) a proposal for reasonable deployment, and how to use the cost-effective equipment to realize the intelligent home furnishing sensor deployment and implementation details; (3) we put forward the new ALD records type, which reconstructs the original polynomial sensor and activity dataset into a larger sparse binary type coordinate with time line. In this way, data gathered from sensors will be easily recorded, and more importantly, computational complexity was reduced and coarse-grained methods enabled, at the expense of tolerable larger occupied memory; (4) for the experiments, we used open datasets for a theoretical accuracy test, and actually deployed our prototype to verify the scheme; (5) finally, we found a suitable way for real-time reaction by a private smart space with high accuracy in a home setting via a heterogeneous wireless network. The prototype can be applied in an embedded device based on TI-cc2530 and Raspberry Pi, which is compatible with wearable devices and smart appliances, integrated with a mobile app for remote surveillance and to recognize the behavior of the residents. The remainder of the paper is organized as follows: in Section 2, we introduce the related work about the design and applications in a smart home. In Section 3, we describe the theoretical 30271 Sensors 2015, 15, 30270-30292 fundamental hypothesis of the behavior routine, and the selection of the sensor and the hardware devices. Subsequently in Section 4, the prototype, heterogeneous architecture process is described in detail. Section 5 gives a full implementation and results to verify our design with home health care applications and mobile applications. Finally, we summarize the discussions and give some conclusions in Section 6. Related Works Recognizing human activities from video is one of the most promising applications of computer vision, and reference [17] has elaborated on the forefront of the advances in the field. Robustness, real-time performance, high processing and storage costs, and intellectual challenges are the main difficulties. In this paper, however, we focus on binary sensors which collect ADLs' streaming sensor records and human activities. Reference [18] described the design and collection methods for ADLs, and we modified the scheme to construct our wireless solution. van Kasteren implemented the temporal probabilistic hidden Markov model (HMM) and conditional random fields (CRF) algorithms [12] to recognize activities from sensor readings, dividing time slices and labels for each. MIT provided three datasets called the PlaceLab Intensive Activity Dataset [8] with more fine-grained algorithms for accurate identification and achieved higher accuracy, but this kind of environment seems perfect yet difficult to deploy in a real home. In the CASAS project [11], the AI-Lab of Washington State University provides several activity datasets and made progress on partnership learning using this CASAS dataset [15] and a web based simulator named Persim for synthesis applications. The main smart homes and ADLs research projects and websites are listed as follows: ' MIT House_n: HttpSeveral research projects have proposed a home energy management system (HEMS) [27], or cost-effective ecosystem to reduce power consumption [28] , and improve the demand response system to accommodate user preference changes and satisfaction. Chen [29] proposed a remarkable Digital Signal Processor and Wi-Fi framework to monitor environment information to refer to, without human behavior detection. In China, taking the average economic level and the amount of population into account, there are 0.2 billion people over 60 this year with a gene coefficient of 4.7. Besides, the design structure of real estate is different from thst used in Western countries, and the popularity of renewable energy generating equipment is insufficient to deploy theoretical cost-effective models. Research on a systematic methodology for smart spaces has been provided [30] , and the quality of experience refers to the factor of human impact on design and perception, experience and expectations of the whole performance [31, 32] . Despite the various smart home applications that have been developed, most of them are application specific and lack a systematic design method. Fortunately, while mobile Internet applications, IPTV, and wearable smart devices are varied and isolated, but most of them adopt a standard communication protocol like Bluetooth, ZigBee or Wi-Fi. Therefore, an effective, reliable and user-friendly design related to ubiquitous computing,
doi:10.3390/s151229797 pmid:26633424 pmcid:PMC4721718 fatcat:ogrx2eebx5hihn5bt3sxtihfli