单对立色流运动下的抠图跟踪

当目标与周围背景相似性较高,且目标运动形式复杂时,很难精确跟踪目标。针对传统目标跟踪方法因目标外观的不精确建模致使模型退化而产生漂移的问题,提出了一种单对立色流特征下的抠图跟踪方法。首先,根据彩色视频帧包含的丰富颜色信息,将单对立色颜色编码方式与Lo G兴趣点检测器结合,获取颜色兴趣点作为目标表示集合;其次,在目标运动预测阶段,通过估计相邻帧兴趣点的流匹配关系,获得预测帧的前景目标兴趣点估计;最后,利用获得的前景兴趣点估计进行抠图,重新划分当前帧的前背景标记,并对模型进行更新。该算法在Segtrack视频跟踪数据集上进行验证,定性定量分析了跟踪目标快速运动,目标形变和光照干扰下的跟踪效果。实验结果表明该算法可有效提高形变目标跟踪的准确性,优于当前目标跟踪的先进算法。

Matting tracking with flow motion in single opponent color space

The main difficulties in object tracking are mainly from the similarity between the target and surrounding background, and the complex situation of object movement. Aiming at the drifting issue in object tracking with classical methods caused by the model degradation due to the inaccurate modeling of the object appearance, a matting based object tracking method under single opponent color flow characteristic is proposed. First, according to the sufficient color information contained in color video frames, a new interest point detector is designed, which combines the visual cortex color surface coding theory (i.e., single opponent (SO) color encoding method) and the Laplacian of Gaussian (LoG) interest point detector to extract the interest points with rich color as the object representation set. Secondly, the foreground object interest points of the predictive frame is estimated through estimating the flow matching relation between two consecutive frame interest points in object motion predicted stage. Finally, the obtained foreground interest point estimation is used for matting, the foreground and background labels of the current frame are generated, and the tracking model is updated. The proposed algorithm was verified on the Segtrack video tracking dataset; both qualitative and quantitative analyses were conducted on the tracking effects of fast motion of the tracking object, object deformation and illumination change. Experiment results demonstrate that the proposed algorithm can effectively improve the accuracy of the deformation target tracking and is superior to the state of the art methods.