Abstract: A precise object tracking algorithm for fusion image based on local image matting was proposed in this paper to improve the tracking effect. Firstly, the rough region of the moving target could be captured by the minus of the sequence’s first two frames and therefore the rectangle was generated for local matting and the discriminative color set of foreground and background. Then the coarse region and the following tracking results automatically provided sufficient and accurate scribbles for matting, which made matting applicable in a tracking system. Finally accurate boundaries of the target could be obtained from matting results so that the model was successfully updated. For experimental image sequences, the mean error of the proposed algorithm was 0.9 pixels between forecasted center and real center of target, and that of traditional mean-shift was 5.2 pixels. The experimental results show that the proposed algorithm can detect the contour of target accurately, ovviously avoids model drift and promotes precision of tracking.

Key words: image matting, image fusion, object tracking, model update, drift

摘要： 为了提高跟踪精度，提出一种基于局部抠像的融合图像全自动精确跟踪算法。首先采用帧间差分法获取运动目标的大致区域，并自动生成局部抠像框，由此采集目标、背景代表颜色集合；在此基础上自动生成抠像所需的草图，实现对目标的抠像；最后对抠像产生的前景映射图进行边缘检测即可获取精确的目标轮廓，并可根据跟踪结果对模型进行更新。对于实验的图像序列，与目标实际中心相比较，抠像跟踪误差均值为0.9像素，传统均值漂移跟踪误差均值为5.2像素。实验结果表明，该方法跟踪结果能完整、清晰地表示目标轮廓，很好地解决了跟踪中的漂移问题，提升了跟踪精度

关键词: 抠像, 图像融合, 目标跟踪, 模板更新, 漂移