Implementasi Algoritma Asosiasi Data Berbasis Batasan Struktural Pada Online Multi Object Tracking

Naufaldi, Dandy (2020) Implementasi Algoritma Asosiasi Data Berbasis Batasan Struktural Pada Online Multi Object Tracking. Undergraduate thesis, Institut Teknologi Sepuluh Nopember.

[img] Text
05111640000011-Undergraduate_Thesis.pdf - Accepted Version
Restricted to Repository staff only

Download (2MB) | Request a copy

Abstract

Object tracking merupakan proses pengenalan identitas objek lintas frame video. Proses ini banyak diterapkan dalam satu alur dengan proses object detection yang menghasilkan informasi bounding box objek. Dari object tracking, dapat diperoleh informasi lintasan pergerakan objek. Kemajuan teknologi membuat perangkat perekaman video menjadi semakin kecil bahkan cukup dengan menggunakan ponsel. Perangkat perekaman tersebut memudahkan pengguna dalam merekam video kapanpun dan di manapun. Namun, rekaman video yang dihasilkan banyak mengalami masalah dari pergerakan perangkat perekam sehingga video menjadi tidak stabil. Pergerakan tersebut dapat berdampak negatif terhadap hasil proses object tracking karena posisi objek yang tidak stabil sehingga asosiasi data antara frame menjadi sulit dilakukan. Pada tugas akhir ini, penulis akan mengimplementasikan algoritma asosiasi data berbasis batasan struktural pada online multi object tracking yang diusulkan oleh Yoon dkk. Pada algoritma ini, batasan struktural digunakan untuk mendapatkan posisi objek relatif terhadap objek lain sehingga dapat mengurangi dampak pergerakan kamera. Program object tracking yang dibuat terintegrasi dengan Kalman filter untuk mendapatkan prediksi posisi berdasarkan informasi posisi objek sebelumnya. Implementasi algoritma ini diujikan pada dataset 2DMOT15 dan mendapat nilai MOTA 20,4% pada data latih. =================================================== Object tracking is a process to recognize object's identity across frame in a video. This process has been implemented as a pipeline with object detection process which produce bounding boxes information. From the output of object tracking, we can get object's movement track. Technology advances have enabled the existence of smaller video recording devices even a phone can give good recordings. Those recording devices enable users to record video anytime and anywhere. But, the video suffer a lot from the movement of the recording device which cause the video to be unstable. The movement might give bad impact to the result of object tracking because unstable object's position make data association process interframe hard to be done. In this thesis, the author will implement a data association algorithm based on structural constraint for online multi object tracking proposed by Yoon et.al. In this algorithm, structural constraint is used to determine object's position relative to other objects which help reduce the effect of camera movement. The program will be integrated with Kalman filter to help predict current object position based on past information. This implementation has been tested on 2DMOT15 dataset and achieved MOTA score of 20.4% on training data.

Item Type: Thesis (Undergraduate)
Uncontrolled Keywords: asosiasi data, batasan struktural, dataset 2DMOT15, Kalman filter, online multi object tracking
Subjects: T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques
Divisions: Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > (S1) Undergraduate Thesis
Depositing User: Dandy Naufaldi
Date Deposited: 07 Aug 2020 02:38
Last Modified: 07 Aug 2020 02:38
URI: http://repository.its.ac.id/id/eprint/77161

Actions (login required)

View Item View Item