Wardhani, Rina Wijaya Kusuma (2018) Pengenalan Aktivitas Manusia pada Video Menggunakan Fitur Matriks Kovarian. Undergraduate thesis, Institut Teknologi Sepuluh Nopember.
Preview |
Text
05111440000021-Undergraduate_Theses.pdf - Accepted Version Download (2MB) | Preview |
Abstract
Saat ini, pengenalan aktivitas manusia pada data video merupakan persoalan yang cukup menantang di bidang visi komputer. Beberapa alat yang dapat digunakan untuk membantu pengenalan aktivitas manusia antara lain sensor Accelerometer, Gyroscope, Camera, dan GPS. Beberapa pengaplikasian pengenalan aktivitas manusia digunakan dalam sistem keamanan atau sarana hiburan.
Pada tugas akhir ini akan dilakukan pengembangan sistem pengenalan aktivitas manusia pada video yang berasal dari Closed-Circuit Television (CCTV) di Departemen Informatika ITS Surabaya. Sistem ini menggunakan fitur matriks kovarian dari fitur optical flow dimana fitur lokal ini dapat menangkap gerakan dinamis sebagai karakteristik dari aktivitas manusia pada video. Pengunaan fitur kovarian dilakukan untuk mengurangi jumlah bag of feature vector hasil dari ekstraksi fitur optical flow menjadi jauh lebih kecil namun tetap dapat merepresentasikan aktivitas manusia. Klasifikasi yang digunakan dalam sistem ini adalah Nearest Neigbour Classifier. Secara umum, pengembangan sistem ini dilakukan dalam lima tahapan proses yakni preprocessing, pengumpulan fitur optical flow, perhitungan matriks kovarian, perhitungan matriks log kovarian, dan yang terakhir klasifikasi jenis aktivitas.
Uji coba menggunakan dataset CCTV menunjukkan bahwa metode yang digunakan pada tugas akhir ini memberikan hasil terbaik pada panjang frame L = 15, overlapping frame P = 1, dan pengambilan data pada waktu malam hari dengan nilai accuracy sebesar 95.86%, rata-rata recall sebesar 96.29%, dan rata-rata precision sebesar 96.01%, serta masing-masing hasil recall dan precision pada tiap kelas aktivitas sebesar 93.57% dan 94.60% untuk aktivitas berlari, 96.39% dan 96.24% untuk aktivitas berjalan, serta 98.91% dan 97.20% untuk aktivitas melambaikan kedua tangan.
=====================================================================================================
Currently, human activity recognition in video data are quite challenging in the field of computer vision. Some tools that can be used to help human activity recognition include Accelerometer, Gyroscope, Camera, and GPS. Many applications of human activity recognition in media systems or entertainment facilities.
In this final project, the development of human activity recognition system will be implemented on video from ClosedCircuit Television (CCTV) in Informatics Department ITS Surabaya. This system uses the covariance matrix feature of the optical flow feature which is the local features that can be used in the human video process. The use of covariance matrix features is done to reduce the number of bags of feature vectors so it becomes much smaller but still can represent human activity. The classification used within this system is the Nearest Neigbour Classification. In general, the development of this system is done in the process of pre-processing, optical flow flow, matrix calculation, covarian log matrix calculation, and activity classification.
Trials using CCTV dataset shows that the method used gives the best result in the use of frame Length L = 15, overlapping frame P = 1, and data collection time on night with an accuracy of 95.86%, average recall of 96.29%, and the average precision of 96.01%, with each recall and precision result in each activity class were 93.57% and 94.60% for running activity, 96.39% and 96.24% for walking activity, 98.91% and 97.20% for waving both hands activity.
| Item Type: | Thesis (Undergraduate) |
|---|---|
| Additional Information: | RSIf 006.3 War p-1 |
| Uncontrolled Keywords: | Pengenalan Aktivitas Manusia, Video, CCTV, Optical Flow, Matriks Kovarian |
| Subjects: | Q Science > QA Mathematics > QA279 Response surfaces (Statistics). Analysis of covariance. T Technology > TA Engineering (General). Civil engineering (General) > TA1650 Face recognition. Optical pattern recognition. |
| Divisions: | Faculty of Information and Communication Technology > Informatics > 55201-(S1) Undergraduate Thesis |
| Depositing User: | Rina Wijaya Kusuma Wardhani |
| Date Deposited: | 28 Jan 2021 23:04 |
| Last Modified: | 28 Jan 2021 23:04 |
| URI: | http://repository.its.ac.id/id/eprint/53693 |
Actions (login required)
 |
View Item |