为了满足监控视频高压缩比的需求,以及应对海量监控视频码流对传输带宽和存储空间带来的挑战,提出一种基于块的监控视频背景图像建模算法(BSBM).首先对编码图像进行分块处理,块大小在建模过程中根据训练集长度增加而增大; 然后计算块中图像残差的梯度,用于刻画运动物体的边缘特征; 最后通过块替换分类检测和块边界检测来判断当前块是否为待选背景块,从而更新背景模型.基于 BSBM 建立的背景图像,提出长参考编码帧间预测策略,将构造出的背景图像作为全局参考图像使用,进一步提高编码性能.在 RD17.0平台上的测试结果表明,在低延时应用场景下,BSBM算法背景图像主观质量较好; 在YUV的3个方向上,平均BD-Rate增益可以达到19.7%,26.6%和26.4%,编码性能提升显著.%To satisfy the demand of high compression ratio of surveillance video, and to deal with the challenge both of the massive surveillance video streams and the limit of storage space, a block-based surveillance back-ground modeling algorithm is proposed. Firstly, the coding unit is divided into blocks, and the size of block is in-creased according to the length of the training set. Then, the gradient of residual blocks is calculated to describe the edge features of moving objects. Finally, the background model is updated by block replacement classification detection and block boundary detection to determine whether the current block is a background block. Based on the background image generated by BSBM, a long reference coding inter-prediction strategy is proposed, and the constructed background image is used as the global reference image to further improve the coding performance. The test results on RD17.0 platform show that in the low-delay application scenarios, the subjective quality of the background image generated by the BSBM algorithm is well, the average of BD-Rate gain for YUV components is 19.7%, 26.6% and 26.4% respectively, which indicates that the coding performance improves significantly.
展开▼
