基于在线目标跟踪算法的研究

Tracking Algorithm based on Online Target

在线目标跟踪是一个具有挑战性的问题。当运动目标表面发生变化、背景光线剧烈变化、背景物体的遮挡以及运动造成的模糊时,需要建立一个有效的物体外在表示模型。为此,本文利用典型的主成分分析(PCA)与稀疏表示建立了有效的观测模型。将l1最小值引入到PCA重建中,将目标物体表示为主要特征和噪声之和的形式。除此之外,当发生遮挡和运动模糊时,为了减少跟踪漂移现象,该文提出了一种新的更新模型而不是简单的对观测模型直接更新。实验证明,该算法在目标物体发生遮挡以及背景光线发生剧烈变化时,对目标跟踪具有较强的鲁棒性。在评价指标重叠率和中心误差方面,算法跟踪精确度明显提高。

Online target tracking is a challenging problem. When the surface of moving target changes and the fuzzy caused by background object＇s shelter and movement appears,it is necessary to establish an effective external representation model of object. Thus,by taking advantage of the typical principal component analysis（ PCA） and sparse representation,an effective observation model is established. And by introducing L1 minimum value into the reconstruction of PCA,the target object is expressed as the sum of the major characteristic and the noise. In addition,when the shade and motion blur,in order to reduce the tracking drifting phenomenon,a novel model is proposed instead of simply updating the observation model.Experiments show that when the target object is covered and the background light changes dramatically,the algorithm is of strong robustness in target tracking. In overlap rate of evaluation index and center error,the algorithm is significantly improred in tracking accuracy.