Deteksi Ketersediaan Slot Parkir Dari Dua Kamera Yang Overlap Menggunakan Metode YOLOv5 dan Image Stitching

Asy'ari, Misbachul Falach (2023) Deteksi Ketersediaan Slot Parkir Dari Dua Kamera Yang Overlap Menggunakan Metode YOLOv5 dan Image Stitching. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025211036-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 September 2025. Download (6MB) | Request a copy

Abstract

Mencari slot parkir yang kosong sangat sulit, terutama saat jam-jam sibuk di kota-kota besar. Beberapa solusi seperti pemasangan sensor di tanah dapat menyelesaikan masalah, namun masih terdapat kekurangan. You Only Look Once version 5 (YOLOv5) merupakan algoritma deteksi objek dapat mengklasifikasikan slot parkir yang kosong atau terisi mobil dan menunjukkan letaknya dalam citra. Pada penelitian sebelumnya, umumnya deteksi ketersediaan slot parkir tidak menyertakan deteksi area overlap, sehingga pada area overlap tersebut terdapat slot parkir yang terdeteksi dua kali. Penelitian ini mengusulkan metode Image Stitching untuk mendeteksi area overlap pada sistem deteksi ketersediaan slot parkir. Penelitian ini membandingkan kinerja beberapa metode Image Stitching, yaitu ORB, BRISK, dan AKAZE. Dataset yang digunakan adalah citra kamera CCTV dari dataset publik CNRPark. Data tersebut memiliki angle area yang overlap dari dua kamera. Dataset dibagi menjadi data latih dan data validasi yang terdiri dari masing-masing set kamera. Langkah pertama adalah membuat ground truth pada data tersebut, kemudian dilakukan augmentasi geometri pada data latih dan data validasi. Selanjutnya dataset di-training menggunakan enam konfigurasi YOLOv5 yang berbeda. Bobot yang didapatkan digunakan untuk proses testing. Data uji yang terdiri dari dua kamera overlap dideteksi area overlap-nya menggunakan beberapa metode Image Stitching. Citra hasil Image Stitching tersebut dibuat ground truth-nya untuk menghitung nilai mean average precision (mAP)-nya pada proses testing. Hasilnya sistem dapat mendeteksi ketersediaan slot parkir dengan skor mAP terbaik sebesar 95,3% pada pasangan Kamera 1 dan Kamera 2 serta 94,2% pada pasangan Kamera 3 dan Kamera 4. Jumlah total ruang parkir sebelum dan sesudah di-stitched juga dibandingkan dalam penelitian ini untuk menunjukkan keakuratan jumlah slot parkir yang tumpang tindih dibandingkan dengan jumlah sebenarnya
================================================================================================================================
Finding an empty parking space is very difficult, especially during rush hours in big cities. Some solutions such as installing sensors on the ground can solve the problem, but there are still drawbacks. The You Only Look Once version 5 (YOLOv5) is an object detection algorithm that can classify empty or occupied parking slots and indicate their location in the image. In the previous study, generally, the detection of parking slot availability did not include the detection of overlapping areas, as a means to the overlapping area there were parking slots that were detected twice. This study proposed an image stitching method to detect overlapping areas in a parking slot availability detection system. This study compared the performance of several image stitching methods, which are ORB, BRISK, and AKAZE. The dataset used is the CCTV camera images from the CNRpark public dataset. The datasets have overlapping angle areas from the two cameras. The dataset is divided into training and validation data for each camera set. The first step is to create a ground truth on the data, then perform geometric augmentation on the training and validation data. Furthermore, the dataset was trained using six different YOLOv5 configurations. The weight obtained is used for the testing process. The test data consisting of two overlapping cameras is detected by the overlap area using several image stitching methods. The ground truth image resulting from image stitching is used to calculate the mean average precision (mAP) value in the testing process. As a result, the system could detect the availability of parking slots with the best mAP score of 95.3% for the Camera 1 and Camera 2 pairs and 94.2% for the Camera 3 and Camera 4 pairs. The total number of parking spaces before and after being stitched is also compared in this study to show the accuracy of the number of overlapping parking slots compared to the actual number.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Parking space detection, YOLO, Image Stitching, Overlapping camera, Deteksi slot parkir, YOLO, Kamera overlap, Image stitching.
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 > 55101-(S2) Master Thesis
Depositing User:	Misbachul Falach Asy'ari
Date Deposited:	21 Jul 2023 02:23
Last Modified:	21 Jul 2023 02:23
URI:	http://repository.its.ac.id/id/eprint/98707

Actions (login required)

View Item