目的 针对背景和摄像机同时运动情况下的运动目标提取与跟踪,提出一种基于稀疏光流的目标提取与跟踪新方法。方法 首先,利用金字塔LK光流法生成光流图像匹配相邻两幅图像的特征点,依据光流图像中的位移、方向等光流信息初步划分背景和前景目标的特征点;然后利用中心迭代法去除不属于目标运动区域的噪声特征点;最后,通过前N帧图像目标特征点的最大交集得到属于目标的稳定特征点并在后续帧中进行跟踪。对于后续跟踪图像中存在的遮挡问题,引入了一个基于特征点的遮挡系数,运用Kalman预估算法得到目标位置的预测,并且在目标重新出现时能够迅速定位目标。结果 与已有的光流匹配算法相比,本文算法的目标特征点误检率降低了10%左右,成功跟踪率达到97%;引入预估器使得本文算法对有遮挡运动目标也能够实现准确跟踪和定位。结论 本文算法对复杂动态背景下无遮挡和有遮挡的持续运动目标跟踪均具有准确识别定位性能,满足实时要求,适用于缓慢或者快速移动的运动场景目标提取和目标跟踪。

Objective This paper proposes a new method based on sparse optical flow to address the problem of target extraction and tracking in dynamic backgrounds. Method First, the pyramid of the LK optical flow method is used to generate an optical flow image to match the feature points between two images. Second, the feature points are divided preliminarily based on the optical flow information on the displacement and direction of the optical flow image. Third, the center iteration method is applied to remove the noise feature points that do not belong to the target motion area. Fourth, the maximum intersection of the target feature points in the first N frames leads to the stable target points that are tracked in the subsequent frames. In the case of blocked targets in subsequent frames, we apply the Kalman estimation method and introduce a blocked coefficient related to feature points to predict the target location and locate the target quickly upon its reappearance. Result The experimental results prove the capability of the proposed algorithm to accurately locate the target. The false detection rate of the target feature points is reduced by 10%, and the tracking rate reaches as high as 97% even when the target is blocked. Conclusion The proposed method demonstrates excellent performance in meeting real-time requirements in dynamic backgrounds and can be applied to tracking slow- or fast-moving targets in blocked or unobstructed scenes.