基于稀疏表示的目标跟踪新算法

A novel target tracking algorithm based on sparse representation

针对跟踪过程中受到光照、噪声等外界干扰导致的跟踪准确率不高的问题,提出一种基于稀疏表示的运动目标跟踪模型。首先对视频图像进行光照归一化处理,通过小波变换获取不同频率信息的子带,对低频部分采用直方图均衡技术改善光照,并结合加权引导滤波对高频部分进行降噪处理;最后运用时频逆小波变换获取优化后的目标图像。在目标重构阶段,针对传统的贪婪算法在迭代过程中忽略了原子间相互关系的问题,采用带宽排除局部最优正交匹配追踪算法,并引入新的判别条件更新相关集半径以获得更为精确的支撑集,从而减少重构误差。在字典更新阶段,设计了新的监督机制,利用相关集分别对目标与判别模板的相似度进行排序,并选定符合条件的相关集中的原子对其进行替换,以减少误差累积。与其他流行算法的对比实验表明,文中所提算法在准确性,鲁棒性方面均有较好的表现。

In order to solve the problem that the low tracking accuracy caused by external interference such as illumination and noise in the tracking process,a moving object tracking model based on sparse representation was proposed.Firstly,the image was normalized by illumination,and wavelet transform was used to obtain sub-band of different frequencies information.Histogram equalization technology was used to improve the illumination of the low frequency part,and weighted guided filter was used to reduce the noise of the high frequency part.Finally,the optimized target image was obtained by using time frequency inverse Wavelet transform.At the stage of target reconstruction,aiming at the problem that the traditional greedy algorithms ignored the interatomic relationship in the iterative process,bandwidth exclusion local optimization orthogonal matching pursuit(BLOOMP)algorithm was used and a new discriminant condition was chosen to update the correlation set radius to obtain more accurate support set,thus reducing the reconstruction error.At the stage of dictionary updating,a new monitoring mechanism was designed to sort the similarity between the target and the discriminant template by using the correlation set,and to select the atoms meeting the conditions in the relevant set to replace them,so as to reduce error accumulation.A comparative experiments show that the proposed algorithm performs better in accuracy and robustness.