摘要
针对稀疏描述符由于关键点检测不稳定及其误匹配造成跟踪失败的问题,提出了一种利用密度描述符对应实现目标跟踪的算法。该算法通过计算目标在相邻两帧之间的密度描述符流,考虑目标的空间分布特性和描述符的权重,得到目标的运动矢量,获得目标在当前帧中的估计;根据密度描述符的运动矢量与目标运动矢量的关系及其匹配程度,更新密度描述符的权重。在大量测试数据上对所提算法进行实验,并与相关跟踪算法进行比较,定性和定量分析表明:当发生光照、遮挡、姿态变化时,所提算法能够稳定地跟踪目标,跟踪成功率均在90%以上,跟踪误差要小于其他算法。
An object tracking algorithm is proposed to overcome the problem that the sparse local invariance feature descriptor depends on the feature detection and always leads to failure.The proposed algorithm uses the dense descriptors correspondences.The dense descriptor flows of objects between consecutive frames are calculated,and object estimations are obtained by considering the spatial distribution and weights of dense flows.Then the weights of descriptors are updated according to the relation and the matching degree between the descriptor motion and the object motion.Qualitative and quantitative analyses on challenging benchmark image sequences show that the average tracking successes rate of the proposed algorithm is over 90% and the performance of the algorithm is better than the performances of several state-of-art methods that assume the constant brightness and template.
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2014年第9期13-18,共6页
Journal of Xi'an Jiaotong University
基金
国家自然科学基金资助项目(61202339
61203628)
陕西省自然科学基金资助项目(2012JQ8034)
关键词
目标跟踪
密度描述符
稀疏描述符
object tracking
dense descriptors
sparse descriptors