A cost-effective cloud-based system for analyzing big real-time visual data from thousands of network cameras
Thousands of network cameras stream public real-time visual data from different environments, such as streets, shopping malls, and natural scenes. The big visual data from these cameras can be useful for many applications, but analyzing this data presents many challenges, such as (i) retrieving data from heterogeneous cameras (e.g. different brands and data formats), (ii) providing a software environment for users to simultaneously analyze the large amounts of data from the cameras, (iii) allocating and managing significant amount of computing resources. This dissertation presents a web-based system designed to address these challenges. The system enables users to execute analysis programs on the data from more than 120,000 cameras. The system handles the heterogeneity of the cameras and provides an Application Programming Interface (API) that requires slight changes to the existing analysis programs reading data from files. The system includes a resource manager that allocates cloud resources in order to meet the analysis requirements. Cloud vendors offer different cloud instance types with different capabilities and hourly costs. The manager reduces the overall cost of the allocated instances while meeting the performance requirements. The resource manager monitors the allocated instances; it allocates more instances if needed and deallocates existing instances to reduce the cost if possible. The manager makes decisions based on many factors, such as analysis programs, frame rates, cameras, and instance types.
Lu, Purdue University.
Off-Campus Purdue Users:
To access this dissertation, please log in to our