Comparative study of path nomalizations for path prediction

Autor: Shimomura, Yuji, Harano, Tomotaka, Tamaki, Toru, Amano, Toshiyuki, Kaneda, Kazufumi, Uchida, Seiichi
Jazyk: angličtina
Rok vydání: 2008
Zdroj: Proceedings of FCV 2008. :284-289
Popis: This paper disccusses two methods on normalization of a sample path for predicting paths of a pedestrian by using eigenspace. A path of a person is defined as a sequence of successive coordinates of the person over frames and represented the path as a vector with 2M elements of M number of coordinates. A problem of their prediction is that the method is based on subspace. To make a subspace from sample paths, all paths need to be normalized and resampled such that a path vector has the same number of elements. Because different sample paths have different number of frames. In this paper, we apply a normalization method using DP matching and discuss results of two predictions : resampling and DP.
Databáze: OpenAIRE