ABSTRAK Nama : Uus Khusni Program Studi : Magister Ilmu Komputer Judul : Pengembangan Pedestrian Detection untuk Kasus Occluded Pedestrian Pembimbing : Prof. Aniati Murni Arymurthy Dr. Heru Susanto(BRIN) Pada saat ini terdapat banyak algoritma object detection dan multiple-person detection dalam bidang Computer Vision. Investasi besar untuk autonomous car oleh perusahaan besar di bidang otomotif dan IT(Information Technology) semakin banyak menarik minat penelitian pada bidang ini. Dalam pengembangan algoritma pedestrian detection ada beberapa kondisi yang akan mempengaruhi tingkat akurasi di antaranya occlusion, illumination change, motion blur, dan texture variation. Di antara kondisi tersebut occlusion adalah masalah signifikan yang mempengaruhi tingkat akurasi. Berdasarkan kondisi tersebut, dibutuhkan pedestrian detection yang efektif dalam berbagai kondisi occlusion. Dalam penelitian ini pedestrian detection diusulkan dan dibangun dengan deep learning. Metode yang diusulkan dibagi menjadi dua tahap. Tahap pertama pedestrian candidate region detection menggunakan kombinasi Faster-RCNN dengan LSTM. Tahap kedua adalah proses target matching di mana tahap ini akan digunakan kombinasi Scale Invariant Feature Transform (SIFT) dan Color Histogram. Penggunaan metode ini diharapkan akan meningkatkan akurasi dari metode sebelumnya yang dijadikan sebagai baseline. Kata Kunci: Color histogram, Faster R-CNN, LSTM, pedestrian detection, SIFT.

ABSTRACT Name : Uus Khusni Study Program : Master of Computer Science Title : Development of pedestrian detection for occluded pedestrian cases Counsellor : Prof. Aniati Murni A Dr. Heru Susanto M.Sc(BRIN) Currently, there are many object detection and multiple-person detection algorithms in the field of computer vision. Considerable investments in autonomous cars by major automotive and IT (Information Technology) companies are attracting more and more research interest in this area. In the development of pedestrian detection algorithms, several conditions will affect accuracy, including occlusion, changes in illumination, blur, and texture variations. Among these conditions, occlusion is a significant problem that affects accuracy. Based on these conditions, effective pedestrian detection is needed in various occlusion conditions. In this study, pedestrian detection is proposed and built with deep learning. The proposed method is divided into two stages. The first stage of pedestrian detectors uses a combination of Faster-RCNN with LSTM. The second stage is the target matching process, in which a combination of Scale Invariant Feature Transform (SIFT) and Color Histogram will be used. The use of this method is expected to increase the accuracy of the previous method, which is used as a baseline. Keywords : Color histogram, Faster R-CNN, LSTM, pedestrian detection, SIFT.