Abstract | ||
---|---|---|
Real-time logo recognition from a live video stream has promising commercial applications. For example, a sports video website broadcasting a live soccer match could show advertisements of brands when their logos appear in the video. Although logo recognition is a well-studied problem, the vast majority of previous work focuses on recognition accuracy, rather than system efficiency. Consequently, existing methods cannot recognize logos in real-time, especially when a large number of logos appear in the video. Motivated by this, we propose a general framework that converts an offline logo detection method to a real-time one, by utilizing the massive parallel processing capabilities of an elastic cloud platform. The main challenge is to obtain high scalability, meaning that logo recognition efficiency keeps improving as we add more computing resources, as well as elasticity, meaning that the resource allocation is guided by the current workload rather than the peak load. The proposed framework achieves these by balancing workload, elastically provisioning resources, minimizing communication overhead, and eliminating performance bottlenecks in the system. Experiments using real data demonstrate the high efficiency, scalability and elasticity of the proposed solution. |
Year | DOI | Venue |
---|---|---|
2016 | 10.1007/978-3-319-39958-4_37 | Web-Age Information Management, Pt II |
Keywords | Field | DocType |
Real-time streams, Logo detection, Elastic cloud platform | Data mining,Broadcasting,Computer science,Workload,Massively parallel,Logo,Provisioning,Real-time computing,Resource allocation,Logo recognition,Multimedia,Scalability | Conference |
Volume | ISSN | Citations |
9659 | 0302-9743 | 0 |
PageRank | References | Authors |
0.34 | 15 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Jianbing Ding | 1 | 68 | 4.72 |
Hongyang Chao | 2 | 495 | 36.96 |
Mansheng Yang | 3 | 9 | 0.81 |