Exhibition-Area Segmentation Using Eigenvectors

In the information age, people feel overwhelmed by information. Large museums are often overwhelming for first time visitors, especially people with limited time. Without professional assistance in exhibition design and arrangement of objects, this issue becomes impossible for them to narrow down the most significant pieces to see. This paper introduces a systematic approach by collecting the visitor information such as the circulation in exhibition space. Then, we can exploit the visitor information to segment exhibition space inherent in circulation behavior of visitors. The segmentation of exhibition space can be achieved with the eigendecomposition of the covariance matrix of the characteristic vectors obtained from visitor dwell time for each time slot. Eigenvectors take advantage of the capability of showing the (first, second, third, and forth) most important circulation behavior of visitors as well as examining the degree of dominance of their corresponding. We, then, adopted the theory of graph spectra for partitioning the exhibit spaces. In experiments, we applied the segmentation approach to the data set obtained from the virtual and real museums: 36 avatars at the Ritsumeikan gallery in Second Life and 45 real visitors at the MIT museum in order to discovering groups of strongly coherent exhibits. The implications are also discussed in the paper.