Detail View

The Internet of Things (IoT) has emerged recently, and the importance of wireless sensor networks in the IoT environment has increased; thus, sensor platforms that can be used to provide a sensor-as-a-service have been studied widely. The sensor filtering techniques employed on sensor platforms are also becoming important as the number of sensors increases significantly in the IoT environment. In addition, Sensor Registry System (SRS) is a system for registering sensor metadata, which supports seamless semantic processing for sensor data. However, the SRS also requires sensor filtering techniques due to the increased number of sensors. Furthermore, the existing sensor filtering techniques have problems because they are influenced by the availability of mobile resources and the status of the mobile network.
In this dissertation, a path prediction-based sensor filtering method (PP-SFM) is, and path prediction-based SRS (PP-SRS) is developed, which is SRS that uses PP-SFM. First, an architecture is constructed for SRS and PP-SFM to develop PP-SRS, before presenting the detailed processes. Then the three-edge pattern-based path prediction algorithm (TEP-PP) is proposed for sensor filtering. TEP-PP divides segments when users move into road segments and it measures the weights of the road segments according to the GPS position of each user. In this dissertation, both SRS and PP-SFM are implemented and tested in experiments. To implement the SRS, a sensor registry data model is designed and the SRS is then developed as a web application. In addition, to implement the PP-SFM, a path repository data model is designed to develop PP-SFM where the RESTful API is used to connect with the PP-SFM. For the experiments, a GPS collector is developed that collects user GPS points. The weights use for path prediction are measured based on the collected user points.
The proposed method is evaluated by measuring the prediction accuracy, service reliability, and sensor preparation performance. The prediction accuracy is compared for a Markov algorithm and the TEP-PP algorithm, where the evaluation demonstrates that TEP-PP obtains higher accuracy. In the service reliability evaluation, simulations are performed with SRS, SRS with SFM, PP-SRS with the fast and close-range prediction algorithm, and PP-SRS with TEP. The simulator randomly generated access failures and then it determines whether the services are successful or not. The results show that PP-SRS with TEP delivered the highest service reliability. In the performance evaluation, the sensor preparation performance denotes the preparation time from receiving the sensor metadata process until semantic interpretation. The preparation time for PP-SRS is longer than that for SRS during the early phase, but the preparation time is shorter than SRS after a threshold value. According to the qualitative evaluation, PP-SFM is limited in terms of domain generality, but it has the advantages of fast service reaction, high compatibility, and high network adaptability in real-time.

1. Introduction	1
1.1 Background	1
1.2 Motivation and Purpose of the Research	3
1.3 Taxonomy	6
1.4 Organization of the Dissertation	8
2. Related Work	9
2.1 Sensor Registry System	9
2.2 Sensor Searching and Sensor Filtering Method	12
2.2.1 Similarity-based sensor filtering	12
2.2.2 Content-based sensor filtering	13
2.2.3 Context-aware sensor filtering	15
2.3 Path Prediction Algorithm	17
2.3.1 Individual behavior pattern-based algorithm	17
2.3.2 Collective behavior pattern-based algorithm	19
2.3.3 Comparison of the path prediction algorithms	21
3. Problem Statement and Solution Approach	24
3.1 Problem Statement and Solution Strategy	24
3.2 Path Prediction-based Approach	28
4. Path Prediction-based Sensor Registry System	32
4.1 PP-SRS Architecture	32
4.2 Sensor Filtering Process	37
5. Path Prediction-based Sensor Filtering Method	40
5.1 Preliminary Information	40
5.1.1 Symbols and definitions	40
5.1.2 Road representation	42
5.1.3 Trajectory and path	43
5.1.4 Time feature	44
5.2 Path Learning Process	45
5.2.1 Initialization	46
5.2.2 Trajectory detection	47
5.2.3 Path identification	50
5.2.4 Weight measurement	52
5.3 Path Prediction Algorithm	53
5.3.1 Three-edge pattern weight measurement	54
5.3.2 Three-edge pattern-based path prediction algorithm	61
6. Implementation and Experiments	64
6.1 Implementation of SRS	64
6.1.1 Purpose and requirement of the SRS	64
6.1.2 Sensor registry model	67
6.1.3 SRS implementation results	71
6.1.4 RESTful API for SRS	76
6.2 Implementation of PP-SFM	78
6.2.1 Path repository model	78
6.2.2 PP-SFM implementation result	79
6.2.3 RESTful API for PP-SFM	81
6.3 Experiment	82
6.3.1 User position collection	82
6.3.2 Weight measurement	86
7. Evaluation	87
7.1 Prediction Accuracy Evaluation	87
7.2 Service Reliability Evaluation	90
7.3 Performance Evaluation	95
7.4 Qualitative Evaluation	98
8. Conclusion	101
Bibliography	104	
국문요약	115