You are here

Object Polygonization in Traffic Scenes using Small Eigenvalue Analysis

Authors: 

Naresh Y G, Venkatesh G M, Noel O'Connor, Suzanne Little

Publication Type: 
Refereed Conference Meeting Proceeding
Abstract: 
Shape polygonization is an effective and convenient method to compress the storage requirements of a shape curve. Polygonal approximation offers an invariant representation of local properties even after digitization of a shape curve. In this paper, we propose to use universal threshold for polygonal approximation of any two-dimensional object boundary by exploiting the strength of small eigenvalues. We also propose to adapt the Jaccard Index as a metric to measure the effectiveness of shape polygonization. In the context of this paper, we have conducted extensive experiments on the semantically segmented images from Cityscapes dataset to polygonize the objects in the traffic scenes. Further, to corroborate the efficacy of the proposed method, experiments on the MPEG-7 shape database are conducted. Results obtained by the proposed technique are encouraging and can enable greater compression of annotation documents. This is particularly critical in the domain of instrumented vehicles where large volumes of high-quality video must be exhaustively annotated without loss of accuracy and least man-hours.
Conference Name: 
The Irish Machine Vision and Image processing (IMVIP) conference
Proceedings: 
https://imvipconference.github.io/
Digital Object Identifer (DOI): 
10.xx
Publication Date: 
31/08/2020
Conference Location: 
Ireland
Institution: 
Dublin City University (DCU)
Open access repository: 
Yes