Multi-feature Hashing Tracking

Chao Ma

doi:10.1016/j.patrec.2015.09.019

Multi-feature Hashing Tracking

2016-01-01

15

PRE

AI

DOI：10.1016/j.patrec.2015.09.019

Original

Original request for help

Save

Abstract

En 中文

Visual tracking is a popular topic in computer vision due to its importance in surveillance, action recognition and event detection. The feature to describe the visual object is an essential element of the tracking model. But there does not exist such kind of feature to handle all situations. Based On this fact, researchers propose the fusion technique to capture robust representation of the object by integrating different features. However, general fusion methods are hard to be applied to tracking algorithm due to the reason of processing speed and online update. To solve this problem, an effective fusion-based hashing method is proposed. The hashing method fuses different features to generate compact binary feature, which could be efficiently processed. In addition, 2D manner and online update model are used to improve the trackers performance. Experimental results demonstrate that our tracker out-performs the state-of-the-art trackers in tested sequences. (C) 2015 Elsevier B.V. All rights reserved.

Keywords:

Hashing

Multi feature

Tracking

Fusion

Multi-feature Hashing Tracking

Abstract

Journal

Organization