IDR Logo

Please use this identifier to cite or link to this item:
Title: A novel point-line duality feature for trajectory classification
Authors: Saini R.
Roy P.P.
Dogra D.P.
Keywords: Fusion
Hidden Markov model (HMM)
Point-line duality (PLD)
Trajectory classification
Issue Date: 2019
Citation: 1
Abstract: Trajectory classification is important for understanding object movements within the surveillance area. Raw trajectories are represented by object location in form of (x,�y) coordinates. The length of trajectories varies in terms of number of points; thus, it is difficult to classify them into correct classes. The raw features extracted from trajectory do not yield satisfactory results in classification. Thus, robust features are needed that can efficiently represent trajectory sequences and help to improve the classification performance. In this paper, we present a new feature vector that is based on the concept of point-line duality (PLD) transformation, i.e., transformation of a trajectory point from its primal plane into a straight line in dual plane. Classification has been done using hidden Markov model (HMM) framework. We also propose a fusion approach combining classification results obtained from raw feature and PLD feature to improve the performance. Experiments have been carried out on raw trajectories with reduced lengths as well as adding Gaussian noise. Proposed approach has been tested on three publicly available datasets, namely T15, MIT, and CROSS. It has been found that the PLD feature outperforms existing features as well as raw feature when used in HHM-based classification. We have obtained encouraging results by feature combination at the decision level with 97, 96.75 and 99.80% accuracy, respectively, on T15, MIT, and CROSS datasets. � 2018, Springer-Verlag GmbH Germany, part of Springer Nature.
Appears in Collections:Research Publications

Files in This Item:
There are no files associated with this item.

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.