제목

최적화 색채 벡터와 Meanshift 기반의 물체 추적

저자명

안효위, 한헌수, 한영준

초 록

Abstract—Mean Shift tracker is a widely used tool to locate the target object robustly and quickly in the sequential image clips. In past Mean shift tracking works, reference histogram and candidate histogram were constructed by uniform partition of original RGB space. This kind partition arranges the data distribution averagely. Compact relationship among immense color statistics is reduced. Uniform distribution also results in final histograms congested with a lot of empty bins which blur the computation cost of the limited memory. In order to reduce such
redundant-empty bin capacity, we present a new Optimal Color Vector in the Mean Shift process for tracking objects. In the proposed mechanism, Optimal Color Vector (OPC), or in other word—the important optimal color, is extracted by closing Euclidean distance clustering which happens inside the original R G B color-3 dimensional spatial domain. The novel OPCs can substitute for original color data. So the target histogram in the candidate frame ROI is mapped by the constructed optimal color clusters and the cluster Indices. In the final, Mean Shift algorithm gives a performance under the new OPC distribution. Comparison in the same situation between our algorithm and conventional Mean Shift algorithm shows that our procedure has a certain advantage on the computation cost. Index Terms—Optimal

원문 수록처 제 7회 로봇종합학술대회 (KRoC 2012), Page 212 ~ 218
자료유형 학술대회자료