摘要
针对运动目标在被遮挡和目标纹理变化大时会导致跟踪丢失以及跟踪误差大等问题,提出了一种改进的压缩感知(CS)算法。算法采用设置Sigmoid函数响应阈值,判定是否存在遮挡,以决定是否更新分类器参数,使得目标在遇到较大遮挡时目标模型不会被错误更新;针对特征单一导致跟踪不稳定问题,提出根据设定融合规则进行灰度特征和纹理特征融合的方法,使得两种特征指导跟踪。实验证明:改进后的算法比传统算法跟踪成功率提高了17.84%,平均误差率降低11.59%。
Aiming at the problems of tracking failure and large tracking error in the case of target occlusion and textures change in moving targets tracking, an improved compressive sensing（CS） algorithm is proposed. Response threshold of a sigmoid function is used to judge whether occlusion exist, to determine whether to update classifier parameter or not, in order to avoid object model is mistake updated when heavy occlusion exists. To solve the tracking instability cause by single feature, a feature fusion method under rules that combined grey features and texture features is presented. Tracking is guided by the fused feature. Experiments verify that the improved algorithm has 17. 84 % higher success tracking rates, and 11.59 % lower average error rate compared with traditional algorithm.
出处
《传感器与微系统》
CSCD
2016年第6期120-123,共4页
Transducer and Microsystem Technologies
基金
天津市教委科技发展基金资助项目(20090718)
关键词
压缩感知
目标跟踪
目标模型
特征融合
compressive sensing（CS）
target tracking
object model
feature fusion
