Title

MULTILEVEL OBJECT TRACKING ON BIG GRAPH DATA USING INTERVAL TYPE-2 FUZZY SYSTEMS IN WIRELESS MULTIMEDIA SENSOR NETWORKS

Abstract

Abstract

Wireless multimedia sensor networks (WMSN) are the key elements of automation
systems applied in different domains from monitoring of an industrial site to home
security for detecting unexpected visitors, from traffic control in a metropolis to immigrant surveillance at a border station. In most of the applications, sensor data needs
to be processed to reveal the information hidden as patterns or data sequences which
is called data analytics. In surveillance applications, this is more crucial since tracking objects in a highly sensitive area or monitoring border for illegal activities are
mission critical. However, interpretation of raw sensor data and unveiling the information inside remains a challenging issue from many aspects. As the measurement
interval of the sensor data is frequent, data needs to be treated as big data because of
the volume and velocity. Unfortunately, traditional approaches don’t perform well in
big data analytics, especially in extracting the complex relationships between data.
In this dissertation, a novel fuzzy object tracking approach which is developed using
a graph-based big data model is proposed by utilization of a multilevel fusion. This
approach consists of three main steps: intra-node fusion, inter-node fusion, and object trajectory construction. Intra-node fusion exploits object detection and tracking
in each sensor while inter-node fusion uses spatiotemporal data along with neighbor
sensors. Then, all trajectories from all sensor nodes are integrated using fuzziness
to construct trajectories in the common ground-plane across the wireless multimedia
sensor network. Since uncertainty naturally exists in trajectory data, type-1 and interval type-2 fuzzy logic systems have been studied on the extracted trajectories as well
as for further analytics like trajectory prediction and anomaly detection.
A prototype system was implemented and several experiments conducted to evaluate
the performance of the proposed approach with both synthetic and real world datasets.
GeoLife dataset and Maritime Cadastre dataset were utilized as input of two different
real world use cases to perform experiments. The results show that usage of thirdlevel fusion, which is called object trajectory construction, in addition to inter-node
and intra-node fusions provides significantly better performance for object tracking
in WMSN applications. Furthermore, interval type-2 fuzzy logic utilization improves
performance in both trajectory extraction and analytics

Biography:

Supervisor(s)

Supervisor(s)

CIHAN KUCUKKECECI

Date and Location

Date and Location

2020-08-11;15:00:00-A105

Category

Category

PhD_Thesis