基于多尺度自适应权重的稀疏表示目标跟踪算法

Sparse Representation Target Tracking Algorithm Based on Multi-scale Adaptive Weight

目标跟踪是计算机视觉中的一个重要研究领域,在交通导航、自动驾驶、机器人技术等众多方面有着广泛应用。基于局部稀疏表示的生成式模型算法ASLA的速度快、跟踪准确性高,但是在复杂跟踪环境下,例如目标局部遮挡、目标外观剧烈变化等,往往会丢失目标。文中分析原算法跟踪原理得到了产生目标跟踪丢失的原因。基于ASLA算法,提出了3点改进方法:1)适应跟踪目标区域大小,采用多尺度分块方式,获取互补的目标局部信息;2)在ASLA特征池化过程中根据分块重构误差建模分块自适应权重,以区分不同分块中包含的判别信息,且在多尺度池化特征中引入不同尺度下的目标遮挡信息作为权重;3)在模板更新时,利用最近帧跟踪结果的稀疏表示权重,使更新模板更相似最近跟踪结果,提高了算法的鲁棒性。实验结果表明,该算法在复杂跟踪环境下相比ASLA等具有更高的跟踪准确度,能够实时、准确地跟踪到目标。

Target tracking is an important research field in computer vision.It is widely used in many aspects such as traffic navigation,autonomous driving and robotics.The generative model algorithm ASLA based on local sparse representation is fast and has high tracking accuracy,but it often loses its target in the face of complex tracking environment,such as target partial occlusion and dramatic change of target appearance.This paper analyzes the tracking principle of the original algorithm to get the cause of target tracking loss.Based on the ASLA algorithm,a three-point improvement method is proposed.1)Adaptive tracking of the target area size using multi-scale blocking method to obtain complementary target local information.2)In the feature pooling process of ASLA,block adaptive weight is modeled based on block reconstruction error to distinguish the discrimination information contained in different blocks,and introducing target occlusion information at different scales as weights in multi-scale pooling features.3)When the template is updated,the weight of the latest tracking results in subspace sparse representation is enhanced to make the updated template more similar to the recent tracking results,and improve the robustness of the algorithm.Experimental results show that the algorithm has higher tracking accuracy than algorithms such as ASLA in complex tracking environment,and can track the target in real time and accurately.