Constrained Online Non Negative Matrix Factorization (CONMF) for Visual Tracking

Abstract

Visual tracking is the process of locating a moving object (or multiple objects) over time using a camera. It is one of the most important components in numerous applications such as military, secure control, crime prevention systems, access control and biometric identification etc. of computer vision. In visual tracking, holistic and part-based representations are both popular choices to model target appearance. The former is known for great efficiency and convenience while the latter for robustness against local appearance or shape variations. Non-negative matrix factorization (NMF) is a feature extraction technique. To adjust NMF to the tracking context, sparsity and smoothness constraints are added to the non-negativity. Putting these ingredients together with a particle filter framework, we proposed the tracker, constrained online non-negative matrix factorization (CONMF). It is possible to achieve robustness to challenging appearance variations and non-trivial deformations in real world. In past decade, there has been made progress for problem of recognizing faces under several unfavorable situations such as changing illumination in an uncontrolled environment. While tracking an object, problems such as sparsity and smoothness appear. By evaluating the tracker on various benchmark sequences containing targets undergoing large variations in scale, pose or illumination. Experiments on real video, including both, indoor and outdoor scenes will demonstrate the effectiveness and robustness of the approach that are subjected to occlusion and alteration in addition to scale.