矿井车载视频图像稳像算法研究

Research on vehicle video image stabilization algorithm for mine

  • 摘要: 针对矿井车载摄像系统拍摄的视频因含有前景运动目标及高噪声造成的全局运动矢量估计误匹配率高、实时性较差等问题,提出了一种基于ORB特征匹配与改进粒子滤波的矿井车载视频图像稳像算法。在运动矢量估计阶段,采用ORB算法提取图像特征点;采用基于图像块的连续3帧间差分法,联合时空一致性准则快速剔除前景运动区域的特征点;结合前景标记区域,对特征点位置进行初次筛选,对保留下来的背景特征点进行配准;利用仿射变换模型实现帧间运动矢量的估计。在运动滤波阶段,采用基于估计窗的实时粒子滤波算法滤除抖动分量,获得补偿参数。实验结果表明,该算法有效避免了前景运动目标对稳像精度的影响,且具有较快的处理速度。

     

    Abstract: For high error matching rate and poor real-time performance of global motion vector estimation caused by foreground moving target and high noise in video image captured by mine vehicle camera system, a vehicle video image stabilization algorithm for mine was proposed which was based on ORB feature matching and improved particle filter. In motion vector estimation stage, ORB algorithm is used to complete extraction of image feature points firstly. Then continuous three-frame difference method based on image block is adopted to quickly mark foreground movement area jointing space-time consistency criteria. Combined with foreground mark area, location of feature points is selected for the first time, and retained points are registered. Finally, affine transformation model is used to estimate motion vectors among frames. In motion filtering stage, a real-time particle filter algorithm based on estimation window is used to filter jitter components, and compensation parameters are obtained. The experimental results show that the proposed algorithm can effectively avoid influence of foreground moving target on image stabilization precision, and has quick processing speed.

     

/

返回文章
返回