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。
In order to reduce the computational complexity ofl_1 norm minimization,a visual object tracking algorithm based on sparse appearance model of local structure in DCT is proposed in paper. The proposed algorithm extracts overlapping local image patches near the target area,and calculates the 2 D-DCT of all the local image patches,and obtains their DCT coefficients. The energy concentration characteristic of DCT is adopted to reduce the dimensions of the dictionary and the number of candidate samples,where low pass filtering can suppress noise and occlusion. After cutting the samples and the dictionary is adopted,the local image sparse coding is obtained. Then,the sparse vectors of target area in all small image patches are weighted with fusion so as to get sparse representation values. Finally,the decision-making model is used to obtain the optimal tracking results. Compared with the existing three tracking algorithms,experimental results demonstrated that the tracking performance of the proposed algorithm approaches or exceeds the contrast algorithms,and greatly reduces the computational complexity.
出处
《电视技术》
北大核心
2017年第7期140-146,共7页
Video Engineering