基于局部结构变换域稀疏外观模型的目标跟踪

发布时间：2018-08-29 11:23

【摘要】：[目的]为了降低稀疏表示目标跟踪算法的计算复杂度,[方法]在粒子滤波框架下提出了基于局部结构变换域稀疏外观模型的视觉目标跟踪算法。[结果]该算法在目标区域附近提取重叠的局部图像块,并计算出所有局部图像块的二维离散余弦变换,获得图像块的变换域系数。变换域的能量集中特性被采用来降低字典的维度与候选样本的数量,并且对系数压缩一定的自由度可以抑制噪声与遮挡影响。采用被裁剪的样本与字典获得局部图像块的稀疏编码,然后将当前目标区域中所有小图像块的稀疏向量加权融合得到目标区域的稀疏表示值,并通过决策模型获取最优跟踪结果。与现有三种最新的跟踪算法比较的实验结果表明,[结论]所提算法的跟踪性能接近或超过对比算法,同时大大减小了l_1范数最小化的计算复杂度。
[Abstract]:[aim] in order to reduce the computational complexity of sparse representation target tracking algorithm, a visual target tracking algorithm based on sparse appearance model in local structure transform domain is proposed in the framework of particle filter. [results] the algorithm extracts overlapped local image blocks near the target region, calculates the two-dimensional discrete cosine transform of all local image blocks, and obtains the transform domain coefficients of the image blocks. The energy concentration characteristics of the transform domain are used to reduce the dictionary dimension and the number of candidate samples, and a certain degree of freedom for coefficient compression can suppress the influence of noise and occlusion. The sparse coding of local image blocks is obtained by using the clipped samples and dictionaries, and then the sparse representation values of the target regions are obtained by weighted fusion of the sparse vectors of all small image blocks in the current target region. The optimal tracking results are obtained by the decision model. The experimental results show that the tracking performance of the proposed algorithm is close to or greater than that of the contrast algorithm, and the computational complexity of minimizing the L _ 1 norm is greatly reduced.
【作者单位】：西南技术物理研究所制导与信息技术部;
【分类号】：TP391.41

【相似文献】