Penerapan Deep Convolutional Generative Adversarial Network Untuk Menciptakan Data Sintesis Perilaku Pengemudi Dalam Berkendara
DOI:
https://doi.org/10.25126/jtiik.20231056978Abstrak
Kecelakaan kendaraan adalah salah satu penyebab kematian tertinggi di Indonesia. Salah satu solusi untuk mencegah kecelakaan adalah dengan menggunakan sensor eksternal untuk mendeteksi kondisi jalan. Namun, penyebab utama kecelakaan adalah kelalaian pengemudi ketika mengemudi yang tidak dapat terdeteksi oleh sensor eksternal. Sensor visual dapat mendeteksi perilaku pengemudi di dalam kendaraan. Penggunaan sensor visual memiliki performa yang lebih baik ketika menggunakan metode deep learning. Salah satu metode untuk meningkatkan performa metode deep learning adalah dengan menggunakan data sintesis hasil model generatif sebagai tambahan data. Deep Convolutional Generative Adversarial Network (DCGAN) adalah salah satu model generatif yang menggunakan lapisan konvolusi. DCGAN terdiri dari dua neural network bernama generator dan discriminator yang membentuk hubungan zero-sum game. Generator menerima masukan berupa gambar asli dengan tambahan noise sebagai input proses latih secara unsupervised, menghasilkan gambar sintesis, sedangkan discriminator menerima gambar asli dan gambar sintesis sebagai input dan menghitung keaslian gambar yang selanjutnya digunakan sebagai nilai loss dengan fungsi loss Binary Cross Entropy. Arsitektur DCGAN terdiri dari beberapa transposed convolutional layer dengan batch normalization dan fungsi aktivasi ReLU dan fungsi aktivasi Tanh sebagai output layer pada generator dan beberapa convolutional layer dengan batch normalization dan fungsi aktivasi Leaky ReLU dan fungsi aktivasi Sigmoid sebagai output layer pada discriminator. Dataset yang digunakan pada penelitian ini adalah dataset ISDDS perilaku umum pengemudi yang dikumpulan pada skenario simulasi dengan jumlah dua ribu gambar. Hasil pengujian menemukan bahwa nilai hyperparameter dapat menghasilkan gambar sintesis perilaku pengemudi di dalam kendaraan yang baik dengan nilai FID sebesar 274,16 pada learning rate discriminator pada 0,0001, β1 discriminator pada 0,8005, learning rate generator pada 0,0017, β1 generator pada 0,1138 selama 43 epoch dengan menggunakan optimizer Adam pada generator dan discriminator.
Abstract
Vehicle crash is one of the leading causes of death in Indonesia. One of the solutions to prevent vehicle crash is by using external sensor to detect road condition. Yet, most crash happened because of driver distraction, which is hard to detect using external sensor. Visual sensor can be used to detect driver activity inside vehicle. Visual sensor that uses deep learning method performs well. One way to increase deep learning method performance is by using additional synthesis data made by generative model. Deep Convolutional Generative Adversarial Network (DCGAN) is a generative model that uses convolution layer. DCGAN consists of two neural networks titled generator and discriminator which create zero-sum game relationship. Generator will receive real image with added noise as input of unsupervised training process, creating synthetic image, while discriminator will receive real image and synthetic image as input and calculate the realness of those image which will be used as loss value with Binary Cross Entropy loss function. The Architecture of DCGAN is composed of multiple transposed convolutional layers with batch normalization and activation function ReLU and activation function Tanh as output layer in generator and multiple convolutional layers with batch normalization and activation function Leaky ReLU and activation function Sigmoid as output layer in discriminator. Dataset used in this research is primary dataset of common driver activity collected in simulation scenario with the size of two thousand images. Experiment result shows that DCGAN is able to create good image synthesis of driver activity inside vehicle with FID of 274,16 using hyperparameter consisting of learning rate discriminator at 0,0001, β1 discriminator at 0,8005, learning rate generator at 0,0017, β1 generator at 0,1138 for 43 epochs by using Adam optimizer on generator dan discriminator.
Downloads
Referensi
RAHAYU, 2022. 2020-2021, Kerugian Materi akibat Kecelakaan Capai Rp 246 Miliar. https://money.kompas.com/read/2022/03/23/151100926/2020-2021-kerugian-materi-akibat-kecelakaan-capai-rp-246-miliar-
BAUTISTA, T., MATTHEW, J., KERDEGARI, H., PERALTA, L., & XOCHICALE, M., 2022. Empirical Study of Quality Image Assessment for Synthesis of Fetal Head Ultrasound Imaging with DCGANs. https://github.com/budai4medtech/
BEANLAND, V., FITZHARRIS, M., YOUNG, K. L., & LENNÉ, M. G., 2013. Driver inattention and driver distraction in serious casualty crashes: Data from the Australian National Crash In-depth Study. Accident Analysis & Prevention, 54, 99–107. https://doi.org/10.1016/J.AAP.2012.12.043
BERGSTRA, J., BARDENET, R., BENGIO, Y., & KÉGL, B., 2011. Algorithms for Hyper-Parameter Optimization. Advances in Neural Information Processing Systems 24.
BORJI, A., 2022. Pros and cons of GAN evaluation measures: New developments. Computer Vision and Image Understanding, 215. https://doi.org/10.1016/J.CVIU.2021.103329
DINGUS, T. A., GUO, F., LEE, S., ANTIN, J. F., PEREZ, M., BUCHANAN-KING, M., HANKEY, J., & DESIGNED, J. H., 2016. Driver crash risk factors and prevalence evaluation using naturalistic driving data. 113(10), 2636–2641. https://doi.org/10.15787/VTT1VC7C
GOODFELLOW, I., BENGIO, Y., & COURVILLE, A., 2018. Deep learning. MIT Press. https://books.google.com/books/about/Deep_Learning.html?id=omivDQAAQBAJ
HARI, C. V., & SANKARAN, P., 2021. Driver distraction analysis using face pose cues. Expert Systems with Applications, 179, 115036. https://doi.org/10.1016/J.ESWA.2021.115036
HU, Y., LU, M., & LU, X., 2020. Feature refinement for image-based driver action recognition via multi-scale attention convolutional neural network. Signal Processing: Image Communication, 81. https://doi.org/10.1016/J.IMAGE.2019.115697
KOAY, H. V., CHUAH, J. H., CHOW, C.-O., & CHANG, Y.-L., 2022. Detecting and recognizing driver distraction through various data modality using machine learning: A review, recent advances, simplified framework and open challenges (2014–2021). Engineering Applications of Artificial Intelligence, 115, 105309. https://doi.org/10.1016/J.ENGAPPAI.2022.105309
LU, Y., CHEN, D., OLANIYI, E., & HUANG, Y.. 2022. Generative adversarial networks (GANs) for image augmentation in agriculture: A systematic review. Computers and Electronics in Agriculture, 200, 107208. https://doi.org/10.1016/J.COMPAG.2022.107208
MCDONALD, A., CARNEY, C., & MCGEHEE, D. V., 2018. Title Vehicle Owners’ Experiences with and Reactions to Advanced Driver Assistance Systems. www.aaafoundation.org
NÉE, M., CONTRAND, B., ORRIOLS, L., GIL-JARDINÉ, C., GALÉRA, C., & LAGARDE, E., 2019. Road safety and distraction, results from a responsibility case-control study among a sample of road users interviewed at the emergency room. Accident Analysis & Prevention, 122, 19–24. https://doi.org/10.1016/J.AAP.2018.09.032
OMERUSTAOGLU, F., SAKAR, C. O., & KAR, G., 2020. Distracted driver detection by combining in-vehicle and image data using deep learning. Applied Soft Computing, 96, 106657. https://doi.org/10.1016/J.ASOC.2020.106657
RADFORD, A., METZ, L., & CHINTALA, S., 2015. Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks. 4th International Conference on Learning Representations, ICLR 2016 - Conference Track Proceedings. https://doi.org/10.48550/arxiv.1511.06434
WAGNER, B., TAFFNER, F., KARACA, S., & KARGE, L., 2022. Vision Based Detection of Driver Cell Phone Usage and Food Consumption. IEEE Transactions on Intelligent Transportation Systems, 23(5), 4257–4266. https://doi.org/10.1109/TITS.2020.3043145
WORLD HEALTH ORGANIZATION., 2018. Global status report on road safety 2018: Summary. https://www.who.int/publications/i/item/WHO-NMH-NVI-18.20
WU, Q., CHEN, Y., & MENG, J., 2020. Dcgan-based data augmentation for tomato leaf disease identification. IEEE Access, 8, 98716–98728. https://doi.org/10.1109/ACCESS.2020.2997001
ZHANG, C., LI, R., KIM, W., YOON, D., & PATRAS, P., 2020. Driver behavior recognition via interwoven deep convolutional neural nets with multi-stream inputs. IEEE Access, 8, 191138–191151. https://doi.org/10.1109/ACCESS.2020.3032344
ZHANG, Y. D., & SANGAIAH, A. K., 2022. Cognitive systems and signal processing in image processing. In Cognitive Systems and Signal Processing in Image Processing. Elsevier. https://doi.org/10.1016/C2020-0-02155-9
Unduhan
Diterbitkan
Terbitan
Bagian
Lisensi
Hak Cipta (c) 2023 Jurnal Teknologi Informasi dan Ilmu Komputer
Artikel ini berlisensiCreative Commons Attribution-ShareAlike 4.0 International License.
Artikel ini berlisensi Creative Common Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)
Penulis yang menerbitkan di jurnal ini menyetujui ketentuan berikut:
- Penulis menyimpan hak cipta dan memberikan jurnal hak penerbitan pertama naskah secara simultan dengan lisensi di bawah Creative Common Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) yang mengizinkan orang lain untuk berbagi pekerjaan dengan sebuah pernyataan kepenulisan pekerjaan dan penerbitan awal di jurnal ini.
- Penulis bisa memasukkan ke dalam penyusunan kontraktual tambahan terpisah untuk distribusi non ekslusif versi kaya terbitan jurnal (contoh: mempostingnya ke repositori institusional atau menerbitkannya dalam sebuah buku), dengan pengakuan penerbitan awalnya di jurnal ini.
- Penulis diizinkan dan didorong untuk mem-posting karya mereka online (contoh: di repositori institusional atau di website mereka) sebelum dan selama proses penyerahan, karena dapat mengarahkan ke pertukaran produktif, seperti halnya sitiran yang lebih awal dan lebih hebat dari karya yang diterbitkan. (Lihat Efek Akses Terbuka).