CAJ | 학술논문

针对视觉跟踪中目标表观的复杂变化问题，提出了一种基于关键区域特征匹配的鲁棒跟踪算法。首先对目标模板进行初始化并通过滤波预测得到目标候选；然后采用自适应标记分水岭算法对目标模板和目标候选进行分割以提取关键区域，并利用像素的空间和频率分布特性对关键区域进行多重特征描述；最后通过关键区域的特征匹配得到目标模板与目标候选的匹配关系，由此确定最终跟踪结果并进行模板更新。对目标发生尺度、遮挡、旋转、光照、姿态、复杂背景以及运动模糊等变化的视频序列进行了仿真测试。实验结果表明，所提算法能够有效处理目标表观的复杂变化问题，尤其对目标的部分遮挡、光照变化以及复杂背景等具有较强的鲁棒性。
침대시각근종중목표표관적복잡변화문제，제출료일충기우관건구역특정필배적로봉근종산법。수선대목표모판진행초시화병통과려파예측득도목표후선；연후채용자괄응표기분수령산법대목표모판화목표후선진행분할이제취관건구역，병이용상소적공간화빈솔분포특성대관건구역진행다중특정묘술；최후통과관건구역적특정필배득도목표모판여목표후선적필배관계，유차학정최종근종결과병진행모판경신。대목표발생척도、차당、선전、광조、자태、복잡배경이급운동모호등변화적시빈서렬진행료방진측시。실험결과표명，소제산법능구유효처리목표표관적복잡변화문제，우기대목표적부분차당、광조변화이급복잡배경등구유교강적로봉성。
In order to cope with the complex variation of target appearance during visual tracking ,a robust tracking algorithm based on feature matching of key regions is proposed .Firstly ,it initializes the target model and obtains target candidate through filter prediction .Then ,it extracts the key regions of target model and target candidate using adaptive marker-based watershed algorithm and describes them with multiple features .Finally ,it matches the key regions to get the mapping from target model to target candi-date and calculates the final tracking results to output and update the target model .The proposed algorithm is tested on the video database containing the appearance variation of scale ,occlusion ,rotation ,illumination ,pose ,background clutters ,and motion blur . The experimental results demonstrate that the proposed algorithm can well cope with the complex appearance variation ,especially shows the robustness to the partial occlusion ,illumination and background clutters .