Video surveillance is an essential tool for many security-related applications, and video is also increasingly used in combination with computer vision technologies in other sensing applications. With the growth of installed camera infrastructure, it would be beneficial to enable a video surveillance system to support multiple applications. More recently, a cloud-based video surveillance system provides a promising solution, where many applications could be implemented by accessing surveillance services on the cloud according to their requirements. In this system, a service-oriented visual sensor network plays an important role, which consists of huge number networked cameras and could provide video content to many applications simultaneously. However, such a network presents several challenges, which need to be addressed and considered along with constraints on the network resources, e.g., coverage, bandwidth and energy capacity. In this thesis, we investigate several efficient resource management and optimization problems for the service-oriented visual sensor network.
We mainly consider two fundamental issues: coverage enhancement and energy efficiency in the service-oriented visual sensor network. Coverage is critical for surveillance applications, which concerns how well the Regions of Interest (RoIs) are observed or covered. Since cameras are directional sensors with a limited Field of View (FoV), they can change their FoVs only by rotating orientations once they are installed. Multiple applications may focus on different RoIs, so an important problem is how to enhance the coverage of these RoIs by controlling the camera sensors’ orientations. Previous studies on coverage enhancement either focus on single area coverage or point(s) coverage. However, ignoring those multiple RoIs or simply treating them as points can cause unwanted coverage, resulting in performance degradation. Hence, we investigate a novel Multiple RoIs Coverage (MRC) problem, aiming to maximize the lowest coverage ratio of all RoIs. Energy efficiency is another challenge to be addressed. In the service-oriented visual sensor network, sensors and the other nodes (e.g., relay nodes) may connect with each other through wireless channel, forming a Wireless Ad-hoc Network (WANET). Wireless cameras are always equipped with limited energy capacities, so video transmission may fail if some nodes are overused during video transmission. Scalable Video Coding (SVC) encodes a video stream into multiple layers (i.e., substreams), and some layers could be dropped to support video scalability. Hence, SVC provides a promising solution for efficient video streaming over the WANET. However, existing works generally focused on Quality-of-Service (QoS) improvement for scalable video transmission, while energy efficiency was not sufficiently studied. We firstly investigate an Energy-Efficient Scalable Video Manycast (E2SVM) problem, aiming to maximize the multicast lifetime and satisfy heterogeneous requirements (e.g., resolution and frame rate) of clients. Then, we further study a Multipath Scalable Video Manycast (MPSVM) problem, where multpath routing techniques are used for many-to-many scalable video transmission in the WANET. Beside energy efficiency, the video quality (i.e., the total received video bitrate) at clients needs to be improved as well in the MPSVM problem. In order to solve aforementioned problems, efficient methods along with extensive evaluations are given in this thesis.