Performance evaluation and enhancements in video sensor networks

Abstract

Video Sensor Networks are developed with the aim of advancing the application performance of the traditional sensor networks and creating new applications. For instance, the reliability of surveillance applications is significantly improved with video streams captured from the events. Rather than detecting an intruder with traditional scalar sensors such as audio and magnetic, an image captured from the surveillance area offers a more detailed inspection of the event. However, for building a video sensor network, mounting CMOS cameras on top of the sensor nodes may not be sufficient. In this thesis, in order to clear up this matter, performance tests on the existing sensor network architectures are conducted by simulations designed for surveillance applications. The performance evaluations indicate that there is a need for enhancements in several layers of the OSI stack. Therefore, firstly, the fragmentation support of sensor MAC protocols which are designed for relaying large data units and data streams such as video are investigated. The improved application quality is observed with proper fragmentation support in terms of reduced latency and increased frame rates. The reason behind the improvements are discovered as the decreased control overhead and the adaptive duty cycle mechanisms. Additionally, the fairness issue in the event based applications are investigated. In order to decrease the event reporting latency (mean response time) and to maximize the overall visual information, a fair queueing method based on the least attained service scheduling is proposed. The results indicate that the reporting delay is decreased and the minimum number of frames received from each event is increased. Also this mechanism is applied to multi-path schemes with variable contention windows.