权重系数自适应光流法运动目标检测

为了实现Horn-Schunck光流法权重系数的自适应设定与更新,研究了权重系数对Horn-Schunck光流法的影响规律,提出一种融合模糊C均值(FCM)聚类的权重系数自适应Horn-Schunck光流法。首先,统计不同权重系数下运动目标检测的光流总值变化曲线。然后,以光流总值的最优化为依据,结合两层模糊C均值(FCM)聚类寻找最优权重和基于固定迭代次数Horn-Schunck光流法的收敛点,从而自适应地获取最优权重系数,并将收敛阈值的人工设定转化为光流值的自动寻优。最后,通过标准视频序列进行测试以验证算法的有效性。实验结果表明:相比于其他权重系数值,最优权重估计的光流图像不但运动目标明显而且噪声较少。对运动目标检测的运行时间为0.106 0s,有用比为0.596 9,幅度误差为0.801 1,满足光流法运动目标检测的最优或次优性能。

Moving object detection by optical flow method based on adaptive weight coefficient

To set and update weight coefficients of Horn-Schunck optical flow method adaptively, the influencing rules of weight coefficients on Horn-Schunck optical flow method is researched. An optical flow method based on adaptive weight coefficients and Fuzzy C-Means(FCM) clustering is proposed. Firstly, it computes varying curves of optical flow total values with different weight coefficients. Then, by combining two levels of FCM clusterings, it finds the optimal weight and the convergence point of Horn-Schunck optical flow method based on fixed number of iterations. By which the optimal weight coefficient is obtained adaptively. Finally, the feasibility of the method is verified based on standard video sequence. The result shows that the optical flow images estimated by the optimal weight obtains evident movement targets with little noise as compared with other weight coefficients and its running time is 0.1060 s, useful ratio is 0.5956, and End-point Error is 0.8011. It achieves the best or the next-best performance.