分块多特征自适应融合的多目标视觉跟踪  

Partitioning and multi-feature adaptive fusion multi-object visual tracking

作　　者：施滢[1] 高建坡[1] 崔杰[1] 孙丽娟[1] 陈宇[1] 

机构地区：[1]南京理工大学电光院,江苏南京210094

出　　处：《传感器与微系统》2015年第9期128-131,共4页Transducer and Microsystem Technologies

基　　金：国家自然科学基金资助项目(61201345)

摘　　要：为提高多目标视觉跟踪算法的实时性和稳定性,提出了分块多特征融合的目标跟踪算法。该算法融合底层颜色、纹理和边缘特征信息,以降低单一目标特征算法容易受复杂环境和目标形变的影响。建立分块目标多特征融合直方图模型,引入目标和背景区分度抑制背景分量,并且结合Kalman滤波器进行预测,在发生遮挡时根据置信度最大子块位置获取遮挡目标位置,实现目标稳定可靠的跟踪。实验结果表明:该算法对每帧图像的平均处理时间为36.2 ms,达到实时性的目的,且算法鲁棒性较强。In order to improve real-time and stability of multiple objects visual tracking algorithm,a target tracking algorithm based on fragment and multi-feature fusion is proposed. This algorithm fuses features of bottom layer color,texture and edge,in order to reduce single target feature algorithm is easily influenced by complex environment and target deformation. Build fragment-based target histogram with multi-feature fusion,introduce distinction degree of target and back-ground to restrain background component and combined with Kalman filter to forecast in the case of occlusion,according to the most confidence fragment for keep out object location,the algorithm achieves reliable goal tracking. Experimental results show that average processing time of the algorithm for every frame is 36. 2 ms,and achieve goal of real-time,and algorithm has strong robustness.

关 键 词：视觉传感 多特征融合 多目标跟踪 分块匹配 KALMAN滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]