Cloud Resource Management for Big Visual Data Analysis from Globally Distributed Network Cameras

There has been tremendous growth of visual data available on the Internet in recent years. This data may be used by a wide variety of applications such as enhancing public safety, urban planning, emergency response, and traffic management. Network cameras produce real-time visual data with versatile content. Millions of network cameras around the world continuously stream data to viewers connected to the Internet and some of these cameras produce high quality data. Analyzing and managing such big visual data requires significant computational and storage resources that can be expensive. The computational requirements may also change based on the run-time conditions such as the time of the day and the scene content. Cloud computing is an ideal choice to meet the resource requirements as it offers resources known as instances with different capabilities and price them according to the usage. There are many options when selecting cloud instances (amounts of memory, the number of cores, and geographic locations). The cost of renting a cloud instance varies based on the options. Inefficient provisioning of cloud resources may become costly in pay-per-use cloud computing. The resource provisioning should also consider the visual data quality. Analyzing high quality data is computationally intensive and time consuming, thereby making it challenging to satisfy the cost and performance requirements of the analyses. High quality data may not be essential for meeting the accuracy requirements of the analyses. For example, high quality may not be required to detect a face as opposed to recognizing a face. Therefore it is important to determine the necessary data quality for the analyses. This dissertation examines the different factors involved in cloud resource management for analyzing big visual data from globally distributed network cameras. A resource manager can significantly reduce the analysis cost by selecting the types, locations, and the number of cloud instances while satisfying the performance and accuracy requirements, and adaptively allocating the cloud resources based on the run-time conditions. A content-aware resource provision system is proposed by studying the trade-offs between quality, accuracy, performance, and cost for big visual data analysis. The necessary video quality for two different applications on data from network cameras is determined