基于分块SURF特征提取的图像目标跟踪算法

由于基于特征的目标跟踪需要对前后两帧图像中的目标进行特征匹配,而传统的基于SURF(speeded up robust features)特征的匹配算法存在匹配时间较长,无法满足目标跟踪条件下实时性要求的情况。本文针对此缺点对SURF特征提取提出了具体的分块并行的解决方案,其中包括自适应地设置分块重叠区域,去除冗余特征点和距离门限法去除离散点的处理;同时通过模板的实时更新以及自适应的抗遮挡处理,保证了短时抗遮挡性能。并通过实验,将传统的基于SURF特征的跟踪算法与本文算法在相同条件下进行跟踪误差和运行时间对比;实验表明针对视频中的待跟踪目标,本文算法较基于传统SURF的图像跟踪算法在降低跟踪运行时间的同时保证了跟踪准确度。并通过遮挡实验,说明抗遮挡处理在本文算法中的实用性和必要性。

Image target tracking algorithm based on blocked SURF extraction

Target tracking algorithm based on feature needs the feature matching between the target region in the former frame and the latter frame,but the traditional tracking algorithm based on SURF(speeded up robust features)cannot meet the requirements of real-time due to long matching time.To solve this problem,the specific block parallel solution for SURF extraction is proposed,including adaptively setting the overlap blocked area,removing redundancy feature points and eliminating the discrete feature points by using distance threshold method,and then,the real-time update of template and adaptive shelter-resisting process are used to ensure the algorithm performance.After that,the tracking error and processing time of proposed algorithm are compared with those of the traditional SURF under the same conditions.The experiment indicates that for the tracking target in the video,the proposed algorithm has superiority in tracking error and processing time than the tracking algorithm based on traditional SURF,and the occlusion experiment indicates the necessity and practicability of the shelter-resisting dispose in proposed algorithm.