Abstract: Traditionally, most tracking algorithms use the single feature to describe the object. In view of the insufficiency of traditional tracking algorithms in complicated background, this paper puts forward an object tracking algorithm used by multi-feature fusion and visual saliency. Firstly, this paper adopts a visual saliency mechanism for      manipulating color histogram data to get saliency feature, uses a hybridstrategy that fuses the BRISK (binary robust   invariant scalable keypoints) feature and saliency information to describe the image, and then extracts the object-     foreground model and object-background model. Moreover, the dynamic feature is extracted by the bidirectional optical flow and error metric and is fused with the static feature which is extracted by the adaptive searching mechanism. Finally, based on the data of matching and tracking procedure in the previous frame, this paper evaluates the object’s scale, rotation and center, and obtains the new target location in the current frame. Experiments show that the proposed algorithm can handle the tracking in the complicated background, and adapt to strong illumination changes, partial occasions, fast motion and so on. The high accuracy and robustness of the proposed algorithm are confirmed by compared with the related algorithms, and the effectiveness and real-time of the method is validated.

Key words: visual saliency, BRISK feature, multi-feature fusion, consensus-based matching and clustering

摘要： 针对大多数目标跟踪算法采用单一特征描述目标，在背景区域出现相似的干扰特征时跟踪精确度较低的问题，提出了一种引入视觉显著性的多特征融合的目标跟踪算法。首先，采用视觉显著性机制处理颜色直方图得到显著性特征，再使用混合特征策略融合显著性特征和BRISK（binary robust invariant scalable keypoints）特征，获取目标前景和背景模型；其次，运用双向光流检测和误差度量提取动态特征，并使用自适应搜索机制提取候选目标区域的静态特征，融合动态特征和静态特征；最后，根据匹配算法估算目标跟踪框的自适应尺度及中心，确定目标在当前帧图像中所处的位置。实验结果表明，该算法能够处理强烈光照变化、目标尺度变化、快速运动及部分遮挡等情况下的目标跟踪问题，并实时稳定地获得单目标跟踪结果。

关键词: 视觉显著性, BRISK特征, 多特征融合, 一致性匹配聚类