(For USM Staff/Student Only)

EngLib USM > Ω School of Aerospace Engineering >

Kernerlized correlation filters parameters optimization for enhanced visual tracking

Kernerlized correlation filters parameters optimization for enhanced visual tracking / Ong Chor Keat
Pengesanan visual telah menjadi salah satu komponen yang penting dalam bidang penglihatan computer kerana ilmu pengetahuan dalam bidang ini dapat digunakan dalam pelbagai aplikasi seperti pengimejan dalam bidang perubatan, pengesahan corak, pengawasan video, robot dalam industri, interaksi antara komputer-manusia dan lain-lain. Penyelidik penyelidik telah menjalankan pengesanan visual dengan pelbagai keadah and pengubasuaian juga dicadangkan untuk menanmbahbaikan keputusan pengesanan. Walaupun kebanyakan kaedah telah mencapai keputusan yang memuaskan, tetapi masih wujud beberapa isu yang perlu diutamakan kerana masalahnya tidak dapat diselesai sepenuhnya dan ini telah menjadi cabaran dalam pengesanan visual. Oleh sebab ini, masih tidak berwujudnya kaedah yang dapat mengesan sasaran dengan sempurnanya. Idea utama yang dicadangkan untuk mempertingkatkan keseluruhan keputusan pengesanan adalah menjalankan teknik optimasi pada parameter yang dipilih. Di sini, prestasi dinilai dengan mengunakan kadar pertindihan (OR) dan ralat lokasi (CLE). Untuk mendapat prestasi yang terbaik, kadar pertindihan perlu ditingkatkan ke tahap maksimum dan ralat lokasi perlu dikurangkan ke tahap minimum berbanding dengan algoritma yang didapati dalam awam. Satu optimasi yang mudah digunakan di sini, keputusan yang terbaik akan dipilih bersama dengan nilai parameter daripada jarak yang ditentukan dalam cara kita. Dengan optimasi, kadar pertindihan purata telah ditingkatkan kepada 0.554 dan ralat lokasi purata dikurangkan kepada 19.803 pixels. Oleh, itu, keadah yang dicadangkan telah mencapai prestasi yang diharapkan dari segi ketepatan dan kemantapan di pengesanan visual pada pelbagai video. _______________________________________________________________________________________________________ Visual tracking has become one of the most important components in computer vision as the knowledge in this field can be applied into a wide range of applications in computer vision such as medical imaging, pattern recognition, video surveillance, industrial robot, computer human interaction, etc. A lot of researches have been conducted and many types of state-of the-art methods and modifications such as sparse representation, online similarity learning, self-expressive, spatial kernel phase correlation filter and others are proposed in order to increase the robustness of the tracking. Despite of many methods has been demonstrated successfully, but there are several issues that still need to be addressed. There still have some unsolvable difficulties in which they become a challenging task to track an object effectively and robustly and it will tend to decrease the accuracy of the results and hence. Until now, there are still no perfect algorithm to track the target flawlessly. In order to improve the performance, the main idea proposed is implementing optimization technique on the selected parameters and obtain a better performance. In this research, the tracking is proposed by using the overlap ratio (OR) and center location error (CLE). In our case, our target is to obtain a better accuracy, which is higher overlap ratio and lower center location error than the result from the algorithms available in public. A simple optimization is used in here, where the global best results with respect to the value of the parameters are selected through a range of values defined in our work. Through the optimization, the overall overlap ratio is increased to 0.554 and overall center location error is decreased to 19.803 pixels. Thus, the proposed method had increased the accuracy and robustness of the visual tracking on many of the video sequences.
Contributor(s):
Ong Chor Keat - Author
Primary Item Type:
Final Year Project
Identifiers:
Accession Number : 875007342
Barcode : 00003107257
Language:
English
Subject Keywords:
Visual tracking; video surveillance; computer vision
First presented to the public:
6/1/2017
Original Publication Date:
3/19/2018
Previously Published By:
Universiti Sains Malaysia
Place Of Publication:
School of Aerospace Engineering
Citation:
Extents:
Number of Pages - 82
License Grantor / Date Granted:
  / ( View License )
Date Deposited
2018-03-19 12:47:16.716
Date Last Updated
2020-05-06 16:18:21.02
Submitter:
Mohd Jasnizam Mohd Salleh

All Versions

Thumbnail Name Version Created Date
Kernerlized correlation filters parameters optimization for enhanced visual tracking1 2018-03-19 12:47:16.716