中文摘要：

为解决并构网络因采用主动队列管理进行拥塞控制，无法区分视频流中不同结构和内容的数据包，难以提高视频播出质量的问题，提出一种结合视觉运动特性的可分级视频拥塞控制机制．该机制基于视觉运动特性的视频数据包重要性标记方法，根据当前队列长度及丢包率与队列长度的非线性关系，改进已有的自适应随机早期检测．当网络发生拥塞时，能根据网络状况和数据包的重要性调整丢包策略．以视频质量分析仪PQA600的注意力加权峰值信噪比、差异平均主观评分和注意加权差异平均主观评分作为评价指标．NS-2仿真实验结果表明，该方法能降低丢包率，提高网络吞吐量，改善视频播出质量．

英文摘要：

In current heterogeneous network video transmissions, the active queue management cannot consider the effects of different structures and content packets on the quality of video broadcast. A new visual motion characteris- tics-based scalable video congestion control mechanism is proposed to solve this problem. This mechanism presents a vision motion characteristics-based video packet importance marking method and improves existing adaptive random early detection according to the current queue length and the nonlinear relationship between the packet loss rate and queue length. When congestion occurs, the packet loss policy is adjusted according to network conditions and the importance of video packets. Attention weighted peak signal to noise ratio （ AW-PSNR）, difference in mean opinion score （DMOS） and attention weighted difference in mean opinion score （ADMOS） of PQA600 are used as video quality assessment values. Results from NS-2 simulation platform have shown that the packet loss rate is reduced; both the network throughput and the video broadcast quality are improved in this mechanism.