Packetization and concealment schemes for delivering H.263+ coded video over internet
Date
2002
Authors
Huitika, Thomas Reino
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Real time streaming H.263+ coded QCIF video for non-interactive security surveillance, unicast and multicast over the Internet using the RTP/UDP/IP transport protocol, must use a packetization scheme that has a high payload efficiency and real time error recovery scheme for lost datagrams due to network congestion. Presently every 256 bytes of the coded video stream are encapsulated into a datagram without considering, the payload's efficiency, the network's MTU, and an error recovery scheme for datagram loss. Also a ratio of four P-pictures to I-pictures is used for error resilience without considering the network path and conditions.
The first part of the thesis addresses the encapsulation problem by investigating two existing packetization schemes that have efficient payloads, which are based upon the RFC2425 document. The Internet's datagram loss process is then simulated to study the two existing packetization schemes robustness against lost datagrams. Simulation is achieved by using a Markov chain and measured datagram statistics from a known network path and conditions that affect decoding of non-interactive real time streaming video. Simulation of the datagram loss process also determines the number of 'Decoder Error Concealment' failures at each ratio of P-pictures to I-pictures (five to fifteen) using five sets of datagram statistics.
The second part of the thesis addresses the error recovery problem for datagram loss by using 'Decoder Error Concealment'. 'Decoder Error Concealment' is used since it is independent of the various sub-network channel resources and can be simple to implement. 'Decoder Error Concealment' schemes that motion compensates the previous picture with predicted motion vectors are only considered.
In this thesis a concealment scheme for the picture 's missing odd or even slices is proposed. The proposed scheme uses two techniques for concealment. The first technique copies previous picture macroblocks for missing macroblocks that do not contain a lot of motion. The second technique uses a fast motion estimation algorithm for missing macroblocks that contain a lot of motion. The decision between the two techniques is
determined by comparing a computed threshold value to a set threshold value. Copying is performed when the computed threshold value is less than or equal to the set value. Motion estimation is performed when the computed threshold value is greater than the set value. The fixed threshold value provides flexibility between the two types of techniques used for concealment. The proposed scheme time is compared to the existing schemes by using PSNR concealment result and CPU time.
Comparison between the concealment results, CPU times, and robustness against datagram loss allows the selection of a packetization and the corresponding concealment scheme for the transport of non-interactive real time streaming video over the Internet.