Comparing accuracy of 4 counting methods in ROI via YOLO detection

Gondoprawiro, Eugene Sebastian Adiwena (2020) Comparing accuracy of 4 counting methods in ROI via YOLO detection. Bachelor thesis, Universitas Pelita Harapan.

	Text (Title) Title (10).pdf Restricted to Registered users only Available under License Creative Commons Attribution Non-commercial Share Alike. Download (1MB)
Preview	Text (Abstract) Abstract.pdf Available under License Creative Commons Attribution Non-commercial Share Alike. Download (356kB) | Preview
Preview	Text (ToC) ToC.pdf Available under License Creative Commons Attribution Non-commercial Share Alike. Download (1MB) | Preview
Preview	Text (Chapter1) Chapter1.pdf Available under License Creative Commons Attribution Non-commercial Share Alike. Download (622kB) | Preview
	Text (Chapter2) Chapter2.pdf Restricted to Registered users only Available under License Creative Commons Attribution Non-commercial Share Alike. Download (1MB)
	Text (Chapter3) Chapter3.pdf Restricted to Registered users only Available under License Creative Commons Attribution Non-commercial Share Alike. Download (1MB)
	Text (Chapter4) Chapter4.pdf Restricted to Registered users only Available under License Creative Commons Attribution Non-commercial Share Alike. Download (6MB)
	Text (Chapter5) Chapter5.pdf Restricted to Registered users only Available under License Creative Commons Attribution Non-commercial Share Alike. Download (362kB)
Preview	Text (Bibliography) Bibliography.pdf Available under License Creative Commons Attribution Non-commercial Share Alike. Download (352kB) | Preview
	Text (Appendices) Appendices.pdf Restricted to Repository staff only Available under License Creative Commons Attribution Non-commercial Share Alike. Download (1MB)

Abstract

Vehicle counting helps provide information regarding the traffic conditions at certain time periods. With access to CCTV footage, a program can be created to count vehicles. To create a program that can count vehicles a program that can process videos and an object detection system is required. YOLO is an object detection system that will be used, Darkflow will act as the program that makes use of YOLO. Two algorithms are going to be implemented into Darkflow. The two other algorithms that will be created in this experiment are the, Simple and Mid-Frame algorithm. The Simple algorithm divides the total count in the video by the number of frames to compute the count value. While the Mid-Frame algorithm only takes the count of the middle frame. The purpose of creating two algorithms is to find out which is more suited for counting in terms of accuracy. Two other pre-existing algorithms, Euclidean Distance and Lane will also be compared for. The two pre-existing algorithms uses YOLO as the object detection system. The purpose of this experiment is to better understand the importance of counting algorithms in vehicle count. The Euclidean Distance and Lane algorithm were tested in low-resolution videos and have accuracy range of about 40.8% in a low-resolution video. The four algorithms are tested using a set of videos in a toll area. These videos are low-resolution (320x240) and only have 5 frames each.To aid this experiment, a manual count data is used. This data contains the vehicle counts at a specific timestamp counted manually. After all four algorithms have been tested for, the vehicle counts of each video gets converted to traffic density. Then the percentage error is calculated for each video. The Simple algorithm has a mean percentage error of 32% while the Lane, Mid-Frame and Euclidean Distance has a mean percentage error of 77%. Because all these algorithms have extremely high percentage errors, none of the algorithms are accurate enough to be using the toll traffic video./ Menghitung kendaraan bisa memberikan informasi tentang kemacetan, di tempat dan waktu yang tertentu. Dengan adanya CCTV, sebuah program bisa dibuat untuk menghitung kendaraan. Untuk membuat program tersebut butuh 2 hal, sebuah program yang bisa memproses video dan sistem deteksi objek. Dalam skripsi ini, sistem deteksi objek yang digunakan adalah YOLO, sedangkan untuk memproses video Darkflow akan digunakan sebagai program utamanya. Dua algoritma akan diimplementasikan ke dalam Darkflow. Dua algoritma lain akan dibuat dalam eksperimen ini yaitu algoritma Simple dan Mid-Frame. Simple mengambil nilai rata-rata dalam jumlah kendaraan yang terhitung dalam seluruh video. Sedangkan Mid-Frame mengambil hitungan frame yang tengah di video. Tujuannya adalah untuk cari tau algoritma mana yang paling akurat. Dalam skripsi ini, dua algoritma yang sudah ada sebelumnya yaitu Euclidean Distance dan Lane akan dibandingkan dengan algoritma Simple dan Mid-Frame. Dua algoritma yang sudah dibuat juga menggunakan YOLO sebagai sistem deteksi objek. Alasan mau cari tahu yang mana lebih akurat adalah karena untuk ke depannya hasil penelitian bisa digunakan untuk lebih tahu kepentingan algoritma dalam hasil hitung kendaraan. Algoritma Euclidean Distance dan Lane diujikan dan mendapat hasil akurasi 40.8% dalam video low-resolution. Algoritma-algoritma tersebut akan diuji dengan file video yang sama di dalam daerah tol yang sama juga. Video tol yang digunakan mempunyai resolusi rendah (320x240) dan setiap video ada 5 frame. Setelah mengumpulkan data hitungan kendaraan, nilai-nilai tersebut akan dikonversi menjadi traffic density. Setelah itu nilai traffic density akan dipakai untuk menghitung nilai percentage error masing-masing algoritma. Nilai mean percentage error Simple adalah 32%, dan Lane, Mid-Frame, Euclidean Distance adalah 77%. Oleh itu tidak ada algoritma yang cukup akurat untuk dipakai dalam video tol.

Actions (login required)

View Item