Object Detection via Heterogeneous Feature Clustering

Abstract

Automated object detection schemes are essential in analyzing surveillance videos However for a large surveillance system with numerous cameras supervised object detection methods require a lengthy training process In this paper we propose an unsupervised approach for identifying different foreground objects Foreground objects are extracted from videos and represented by heterogeneous features To assess similarities among feature vectors with unequal lengths a sequence matching procedure is proposed Then different properties of foreground objects are identified by clustering feature vectors in each feature space Finally a tree-structured clustering algorithm is used to identify the foreground objects with similar properties The results of real data set confirm the effectiveness of the proposed algorithm in separating different foreground objects

Date of Award	2014 Jul 28
Original language	English
Supervisor	Pau-Choo Chung (Supervisor)