摘要： 耗时的运动估计运算给实时视频编码带来较大困难，为提高监控视频编码的实时性，提出一种用于嵌入式视频监控系统的快速运动估计算法。该算法根据监控背景相对固定的特点，使用多层提前终止准则，通过运动矢量时空相关性的特性来预测搜索起点，采用改进型的十字搜索模板进行搜索。实验表明，与自适应十字搜索算法相比，该算法在平均峰值信噪比略有下降的情况下，搜索速度得到提升，符合嵌入式视频监控的实时性优先原则。

关键词: 运动估计, 块匹配, 十字搜索, 嵌入式监控, 视频编码

Abstract: Motion estimation of time-consuming brings great difficulties to real-time video encoding. In order to improve monitoring of real-time video encoding, a rapid motion estimation algorithm for embedded video monitoring system is proposed. Based on the characteristics of a relatively fixed monitoring background, the algorithm uses multi-criteria of early stop, predicts search starting point through vector movement relatively characteristics of time and space, finally, uses improved rood search pattern starting search. Experiment shows that the algorithm has better search rate than adaptive rood search algorithm with slight PSNR decrease, matches the real-time priority principle in embedded video monitoring system.

Key words: motion estimation, block matching, rood search, embedded monitoring, video coding

中图分类号: 

• TP311