Abstract: In video target tracking,mismatched feature points tend to cause reduction in the tracking performance.To address the problem,a video target tracking method using Feature Point Mismatch Elimination(FPME) is proposed on the basis of an existing method,GMS,which integrates ORB and grid statistics.The method employs ORB algorithm to ensure the real-time performance of feature point matching in video sequences.Then,a two-stage elimination method is adopted to eliminate mismatched points.In the first stage,the K-means algorithm is used to quickly and roughly eliminate the wildly inaccurate matching relationships,improving the proportion of the accurate matching pairs.In the second stage,the splitting method is used to precisely eliminate the matched pairs with larger deviations,further improving the success rate of matching between target feature points.The experimental results show that compared with GMS,ASLA,HDT and other mainstream algorithms,this method performs better in matching accuracy,speed and other evaluation indicators in cross-frame matching and continuous tracking of video sequences.

Key words: video target tracking, feature point, mismatching, K-means algorithm, splitting method

摘要： 针对视频目标跟踪中因特征点误匹配造成跟踪性能下降的问题，在融合二进制特征描述算法（ORB）与网格统计的视频跟踪方法（GMS）框架下，提出一种基于GMS与特征点误匹配剔除（FPME）的视频目标跟踪方法。利用ORB算法确保在视频序列中特征点匹配的实时性，采用“粗-精”两阶段的剔除方法，即先利用K-means算法快速粗略地剔除误差较大的特征点匹配关系，提高正确匹配对所占的比例，再利用分裂法精确剔除偏离程度较大的匹配对，提高目标特征点之间的匹配成功率。实验结果表明，在视频序列的跨帧匹配与连续跟踪实验中，该方法相对于GMS、ASLA、HDT等当前主流算法在匹配精度、速度等评价指标上都能得到较好的结果。

关键词: 视频目标跟踪, 特征点, 误匹配, K-means算法, 分裂法

CLC Number: 

• TN919