L1-L2范数联合约束的鲁棒目标跟踪

针对稀疏原型跟踪方法中未考虑正交模板系数的密集性的问题,本文提出一种L1-L2范数联合约束的鲁棒目标跟踪。首先,该方法建立基于L1-L2范数联合约束的目标表示模型,对PCA基模板系数和琐碎模板系数分别进行L2范数和L1范数正则化约束,不仅提高了跟踪的准确性,而且保证了对目标遮挡的鲁棒性;其次,针对目标表示模型的优化问题,运用岭回归和软阈值收缩方法快速迭代求解PCA基模板系数和琐碎模板系数;最后以粒子滤波为框架,利用目标未被遮挡部分的重构误差和稀疏噪声项建立观测模型,并结合提出的L1-L2范数联合约束的算法实现目标跟踪。实验结果表明,与5个现有的跟踪算法相比,本文的跟踪算法具有更好的准确性和鲁棒性。

Robust object tracking via L1-L2 norm simultaneous constraint

Aiming at the problem that the density of the orthogonal template coefficients is notconsidered in the sparse prototypes tracking method, a robust object tracking algorithm based on L1-L2 norm simultaneous constraint is proposed. Firstly, the object representation model based on L1-L2 norm simultaneous constraint is established,The L2 norm regularizati on constraint on the PCA basis templates coefficients and the L1 norm regularization constr aint on the trivial templates coefficients are performed,which not onlv improves the tracking accuracy but also guarantees the robustness to target occlusion. Secondly, aiming at the optimization problem of object representation model,the ridge regression and soft threshold shrinkage methods are used to conducta fast iteration and solve the PCA basis template coefficients and trivial template coefficients,respectively. Finally, taking particle filter as the framework, an observation model is established with the reconstruction error of the object unoccluded part and sparse noise term, and the object tracking is realized with the proposed algorithm based on L1-L2 norm simultaneous constraint. The experiment results show that the proposed tracking algorithm achieves better accuracy and robustness in comparison with five state of the art tracking algorithms.