基于结构化判别稀疏表示的目标跟踪

Structured discriminant sparse representation based object tracking

针对稀疏表示目标跟踪算法采用整体模板且区分目标与背景的能力差的缺点,该文提出了一种改进算法。采用尺度不变特征变换(SIFT)对目标进行特征提取。采用结构化稀疏表示的外观模型对候选目标进行稀疏表示,得到稀疏系数。通过正负样本设计并训练判别分类器,然后对候选目标进行分类,获得置信值。采用上一帧的跟踪结果对分类器与字典进行更新。对该文算法进行了仿真研究。计算仿真结果中3种测试序列的平均重叠率和平均中心点误差,Deer测试序列的值为0.633 8和9.397 6,Car11测试序列的值为0.677 5和1.943 3,Caviar2测试序列的值为0.753 5和3.838 2。

An improved algorithm is proposed aiming at such shortcomings of sparse representation based object tracking algorithm as using an overall template and the poor ability of distinguishing targets from a background. Scale-invariant feature transform（SIFT） is used to extract the features of a target. Candidate objects are sparsely represented using appearance models of structured sparse representation,and sparse coefficients are obtained. A discriminant classifier is designed and trained by positive and negative samples,candidate objects are classified,and a confidence value is obtained.The tracking result of the previous frame is used to update the classifier and the dictionary. The improved algorithm is simulated. The average overlap ratio and average center point error of 3 test sequences of the simulation results are calculated,and Deer test sequence＇s are 0.633 8 and 9.397 6,Car11 test sequence＇s are 0.677 5 and 1.943 3,Caviar2 test sequence＇s are 0.753 5 and 3.838 2.