Sidharta, Hanugra Aulia (2025) Prediksi Pola Tingkah Laku Grup Pejalan Kaki Ketika Menyeberang Menggunakan Fitur Kinematik Angular dan Kerapatan Spasial Berbasis Estimasi Pose 2D. Doctoral thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 7022201002_Dissertation.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
7022201002_Dissertation.pdf - Accepted Version Restricted to Repository staff only until 1 April 2027. Download (23MB) | Request a copy |
Abstract
Ketika berjalan di sisi jalan, pejalan kaki mempunyai kecenderungan berjalan dalam kelompok kecil, kelompok ini dapat terbentuk ketika mereka bergerak dengan ritme dan arah tujuan yang sama. Saat berada dalam grup yang sama, antar pejalan kaki melakukan interaksi secara dinamis dengan tujuan menjaga kohesi grup. Terbentuknya grup pejalan kaki dapat diamati ketika mereka sedang berjalan di sisi jalan, maupun saat mereka menunggu di sisi jalan untuk menyeberang. Keputusan menyeberang dapat diambil berdasar keputusan individu atau secara aklmasi dengan mengikuti satu orang yang bertindak sebagai pemimpin grup. Pengamatan pola perilaku pejalan kaki dilakukan dengan menggunakan set data Joint Attention in Autonomous Driving (JAAD). Pengamatan pola perilaku pejalan kaki ketika menyeberang dilakukan memanfaatkan hasil estimasi sendi berbasis pose 2D. Dengan melakukan ekstrasi fitur temporal angular melalui perhitungan sudut kinematika dan fitur kerapatan berdasar posisi spasial sendi. Grup pejalan kaki dapat dideteksi melalui segmentasi semantik berbasis fitur kerapatan beriorientasi pada arah pejalan kaki melalui arsitektur \textit{shallow} U-net, dengan melakukan modifikasi terhadap jumlah lapisan. Metode segmentasi shallow U-net memiliki kinerja yang lebih baik dibandingkan arsitektur U-net dengan mencapai kestabilan yang lebih cepat pada enam epoch pertama.Ketika menyeberang secara berkelompok, pejalan kaki membuat keputusan menyeberang secara konsensus. Keputusan menyeberang ini dapat terjadi dengan mengikuti keputusan salah satu pejalan kaki. Prediksi menyeberang dapat dilakukan melalui model Multi Input Single Output (MISO), dengan kemampuan menerima banyak input dengan satu output berdasarkan keputusan konsensus. Prediksi keinginan pejalan kaki menyeberang juga dapat dilakukan menggunakan hasil estimasi pose kepala. Dengan memanfaatkan pengamatan terhadap nilai roll, pitch dan yaw dari estimasi pose kepala
======================================================================================================================================
When walking on the side of the road, pedestrians tend to walk in small groups, these groups can form when they move with the same rhythm and direction. When in the same group, pedestrians interact dynamically with the aim of maintaining group cohesion. The formation of pedestrian groups can be observed when they are walking on the side of the road, or when they are waiting on the side of the road to cross. The decision to cross can be taken based on individual decisions or acclimatise by following one person who acts as the group leader. Observation of pedestrian behavior patterns was carried out using the Joint Attention in Autonomous Driving (JAAD) dataset. Observation of pedestrian behavior patterns when crossing was carried out using the results of 2D pose-based joint estimation. By extracting temporal angular features through calculating kinematic angles and density features based on the spatial position of the joints. Pedestrian groups can be detected through semantic segmentation based on density features oriented to the direction of the pedestrian through the \textit{shallow} U-net architecture, by modifying the number of layers. The shallow U-net segmentation method performs better than the U-net architecture by achieving faster stability in the first six epochs. When crossing in groups, pedestrians make a decision to cross by consensus. This decision to cross can occur by following the decision of one pedestrian. Prediction of crossing can be done through the Multi Input Single Output (MISO) model, with the ability to accept multiple inputs with one output based on consensus decisions. Prediction of pedestrians' desire to cross can also be done using the results of head pose estimation. By utilizing observations of roll, pitch and yaw values from head pose estimation
| Item Type: | Thesis (Doctoral) |
|---|---|
| Uncontrolled Keywords: | density Feature, group of pedestrian, kinematic Feature, pedestrian crossing prediction, fitur kerapatan, fitur kinematika, grup pejalan kaki, prediksi perilaku pejalan kaki |
| Subjects: | Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20001-(S3) PhD Thesis |
| Depositing User: | Hanugra Aulia Sidharta |
| Date Deposited: | 20 Jan 2025 06:50 |
| Last Modified: | 20 Jan 2025 06:50 |
| URI: | http://repository.its.ac.id/id/eprint/116445 |
Actions (login required)
 |
View Item |