视频分析中利用可变阈值的运动估计算法研究

Research on motion estimation algorithm based on variable threshold in video analysis

研究的是基于阈值的运动矢量估计技术,根据视频序列中运动矢量的特性,构建阈值函数。对于匹配搜索的不同位置自动生成对应的阈值,建立从中心向外逐渐松弛的约束条件,及时终止对不必要候选块的匹配搜索。不仅能够提高匹配速度,而且能够避免遍历搜索中容易陷入局部最小化的问题。同钻石搜索法相结合,能进一步提供运算速度。还提出了通过均值累加阈值法消除由于全局运动补偿引入的噪声,该方法可以确保捕获运动对象的真实性。通过实验验证,对大多数的视频序列,能在不损失精度的情况下,运动估计速度提高几乎一个数量级,运动矢量捕获的真实率在70%以上。

The research of this paper is a technology of motion estimation based on the threshold value. The threshold function is constructed according to the characteristics of motion vector in a video sequence. Different fields can automatically generate the corresponding threshold value. Relaxation constraints can be established gradually from the center outward and the matching of the unnecessary candidate block can be terminated. The method is not only improving the matching speed but also avoiding producing the local minimum problems as well. And it further provides the computing speed by using diamond searching algorithm. This paper also advises the mean cumulative threshold method to eliminate the noise introduced by the global motion compensation and ensures the veracity of the moving object. Experiments prove that the motion estimation speed is increased by almost an order of magnitude without loss of precision. The veracity rate of motion captured is more than 70% for most video sequences.