A Comparative Study of Tracking Moving Objects in videos
Visual tracking is considered to be one of the most important challenges in computer vision with numerous applications such as object recognition and detection. In the present paper, five tracking techniques will be introduced circulant structure with kernels (CSK), Kernelized correlation filters (KCF), Adaptive color attributes (ACT), distractor – awareness tracker (DAT), and Multi-Template Scale KCF (MTSc-KCF) for the visual object tracking (VOT14), and VOT15 challenge datasets. Performance evaluation for each method was calculated using four measures; center location error (CLE), overlap precision (OP), distance precision (DP), and speed in frames per second (FPS). Results have shown that KCF tracker is the fastest technique in VOT14 but the CSK tracker is the fastest in VOT15. They are used in time-critical application with satisfactory performance. MTSc-KCF, and KCF achieve the best results in most sequences and the highest precision at lower threshold. Each tracker performs favorable and competitive results in some sequence and fails in others. So it is noted that the choice of the tracker is application dependent.
Keywords: Visual tracking; correlation filter; distractor; distance precision; precision plot
Download Full-Text
ABOUT THE AUTHORS
Walaa Omar El-Farouk Hussein Badr
I work in Mansoura Higher Institute for Engineering and Technology.
Hossam. El-Din Mostafa
Lecturer of Electronics and Communication of Engineering, Mansoura University
Rasheed Mokhtar El-Awady
Prof. of Electronics and Communication Engineering, Mansoura University. Head of communication and computers, Delta University.
Walaa Omar El-Farouk Hussein Badr
I work in Mansoura Higher Institute for Engineering and Technology.
Hossam. El-Din Mostafa
Lecturer of Electronics and Communication of Engineering, Mansoura University
Rasheed Mokhtar El-Awady
Prof. of Electronics and Communication Engineering, Mansoura University. Head of communication and computers, Delta University.