遮挡情况下目标局部特征在跟踪算法中的应用

为了解决视频目标跟踪中的遮挡问题,本文提出了一种新的目标局部形状特征——仿射不变三角形来对目标进行描述,所提取的仿射不变三角形特征具有平移、旋转和尺度不变的特性,在各种异常情况下仍然能够稳定地表征目标。本文采用仿射不变三角形特征量来进行有效的遮挡状态的判断,并将目标未被遮挡部分的局部形状特征量与运动预测结合起来进行基于模板的匹配跟踪,能够避免遮挡所导致的目标外观数据改变而引起的目标运动状态计算的错误,以解决目标在跟踪中的遮挡问题。实验结果证明,本文提出的局部形状特征能够比较准确地判断目标的遮挡状态,并且能够有效地提高跟踪算法的准确性和稳定性。

Local Features for Occlusion in Object Tracking

To solve the occlusion problem of object tracking in videos,a local shape feature is proposed to estimate the occlusion state.In this correlate matching algorithm,the methods combine the local features which are not occluded and the motion predictions to calculate the object portion.In this way,the influence brought by the occluded is avoided.The experimental results show that the algorithm is robust to solving the occlusion problem,and the occluded state judged by the proposed local shape feature is correct.