|Keywords:||Engineering and Technology; Teknik och teknologier; Master of Science Programme in Computing Science and Engineering; Civilingenjörsprogrammet i Teknisk datavetenskap|
|Full text PDF:||http://urn.kb.se/resolve?urn=urn:nbn:se:umu:diva-99062|
Video Content Analysis (VCA) is usually computationally intense and time consuming. In this thesis the efficiency of VCA is increased by implementing a distributed VCA architecture. Automatic speech recognition is used as a case study to evaluate how the efficiency of VCA can be increased by distributing the workload across several machines. The system is to be run on standard desktop computers and need to support a variety of operating systems. The developed distributed system is compared to a serial system in use today. The results show increased performance, at the cost of a small increase in error rate. Two types of load balancing algorithms, static load balancing and dynamic load balancing, is evaluated in order to increase system throughput and it is concluded that the dynamic algorithm outperforms the static algorithm when running on a heterogeneous set of machines and that the differences are negligible when running on a homogeneous set of machines.