研究了一种基于多线索融合的目标跟踪算法并在TI DM3730上实现.该算法结合在线AdaBoost和颜色目标跟踪算法,选用Haar小波和核颜色直方图两种特征类型,利用粒子的状态散度矩阵的行列式对这两种不同类型的特征进行融合,实现了对所选定目标的鲁棒跟踪.分析及实验结果表明,与传统的基于单一特征的目标跟踪方法相比,本算法融合了两种不同类型特征的互补性,在具有挑战性的复杂环境下,例如目标遮挡,旋转以及光照变化等,都有更好的跟踪效果.%A target tracking algorithm based on multi-cue fusion is investigated and implemented on TI DM3730. The proposed algorithm combines the online AdaBoost and color kernel histogram algorithm together, selects Haar wavelet feature and kernel color histogram as two kernel features, and utilizes the determinant of the divergence matrix of the particle state to fuse these two difference features. With the proposed algorithm, the selected target can be tracked quickly and robustly. Analytical and experimental results show that the tracking efficiency and effect of the algorithm fusing the complementary of two difference features can perform better compared to the traditional single-feature-based tracking methods in most challenging environments, such as occlusion, target rotation and illumination.
展开▼