具有动态弹性稀疏表示的鲁棒目标跟踪算法

Dynamic elastic net sparse representation robust visual tracking

目标跟踪问题中目标所在环境的变化对跟踪效果有较大影响.鉴于此,提出一种基于弹性网结构的稀疏表示模型,并在粒子滤波框架下设计一种应用稀疏表示模型的抗干扰动态弹性网目标跟踪算法.同时,设计一种根据环境变化程度动态更新稀疏表示模型参数的方法,以克服光照变化等干扰对算法跟踪质量的影响.此外,所提出算法通过使用各向异性核函数计算各候选区域为跟踪目标所在位置的概率,能够提高跟踪算法的准确性,并改进字典模板更新方法,确保模板更新的准确性与及时性,保证跟踪质量.经实验验证,所提出的动态弹性网跟踪算法与其他跟踪算法相比,在光照等扰动下具有更好的跟踪效果,在遮挡及快速运动等情况下也能够有效保证跟踪精度.

In visual tracking,the target’s environment has a significant influence on the tracking result.To solve this problem,we propose a sparse representation model based on the elastic net and design an anti-jamming visual tracking algorithm under the particle filter framework.To overcome the influence of light change and other disturbances on the tracking result,we develop a method to dynamically update sparse representation model parameters according to the environment change.Besides,using the anisotropic kernel function to calculate the probability that each candidate region is the tracking target’s location,the proposed algorithm improves the tracking algorithm’s accuracy.Furthermore,we improve the dictionary template updating method to ensure the accuracy and timeliness of template updating and ensure the tracking quality.Experimental results show that compared with other tracking algorithms,the dynamic elastic network tracking algorithm proposed has a better tracking effect under disturbance,such as illumination.Moreover,the algorithm can virtually guarantee tracking accuracy under occlusion and fast motion.