Frame allocation algorithms for multi-threaded network cameras

Abstract

This paper presents a first attempt to solve a challenging problem, proposing novel and successful algorithms to efficiently distribute video frames from network cameras to many concurrent clients. The usual scenario studied is composed of a camera generating video frames at a given rate and distributing them over a network to several concurrent clients. In general, the idea is to allocate one thread per client at the camera, sharing a pool of one-frame buffers. The algorithms studied consider the allocation of buffers to new frames and the allocation of frames to clients. We study different combinations of algorithms, buffers and clients in order to find an optimal solution for the usual scenarios we face when the network camera is under heavy use. The main conclusion is that frame allocation algorithms have a strong impact on system performance: under the same conditions, client performance improves from 4 to 25 frames per second with the best algorithm combination at the camera. © 2010 Springer-Verlag.