DETEKSI OBJEK BAYANGAN KENDARAAN MENGGUNAKAN FASTER R-CNN
DOI:
https://doi.org/10.21009/03.1201.FA04Abstract
Abstrak
Objek bayangan memiliki intensitas dan bentuk yang beragam, yang dapat menimbulkan masalah dalam sistem visi kendaraan otonom. Bayangan yang ditimbulkan dari pohon, bangunan, dan objek lain di sekitar jalan dapat mempengaruhi kinerja sistem pengenalan dan pelacakan target. Maka itu, diperlukan suatu model pendeteksian bayangan untuk mengetahui lokasi bayangan agar dapat digunakan pada penelitian terkat eliminasi bayangan. Penelitian ini bertujuan untuk mengetahui tingkat akurasi model dengan variasi dataset yang kami berikan dan mendefinisikan masing-masing label objek non-shadow dan shadow merupakan metode yang digunakan untuk membedakan antara bayangan dan objeknya yang mirip. Pelatihan model dilakukan dengan fine-tuning Faster R-CNN pada kerangka kerja Pytorch menggunakan arsitektur ResNet50 sebagai rancangan dasar. Implementasi model untuk dapat mendeteksi bayangan diterapkan pada video perjalanan kendaraan otonom. Hasil penelitian menunjukkan bahwa dari kelima model yang dibuat, model P5 berhasil mendeteksi bayangan dengan rata-rata akurasi F1-score sebesar 46%.
Kata-kata kunci: Bayangan, Deteksi, Faster R-CNN, R-CNN, ResNet50, Pytorch
Abstract
Shadow objects exhibit varying intensities and shapes, which can pose problems in autonomous vehicle vision systems. Shadows generated by trees, buildings, and other objects in the vicinity of the road can impact the performance of the recognition and tracking system. Thus, a shadow detection model is necessary to determine the location of shadows, which can be employed in studies related to shadow removal. This study aims to determine the accuracy level of the model with our given diverse dataset and defining distinct labels for non-shadow and shadow objects to differentiate between shadows and similar-looking objects. The model training was performed by fine-tuning Faster R-CNN on the PyTorch framework, utilizing ResNet50 as the backbone architecture. The implemented model aimed to detect shadows in videos of autonomous vehicle. The results indicated that out of the five models developed, P5 model successfully detected shadows with an average accuracy based on F1-score is 0.46%.
Keywords: Shadow, Detection, Faster R-CNN, Faster R-CNN, R-CNN, ResNet50, Pytorch
References
[2] A. Yoneyama et al., “Moving cast shadow elimination for robust vehicle extraction based on 2D joint vehicle/shadow models,” Proceedings of the IEEE Conference on Advanced Video and Signal Based Surveillance, Miami, USA, pp. 229-236, 2003, doi: 10.1109/AVSS.2003.1217926.
[3] B. Chougula et al., “Road segmentation for autonomous vehicle: A review,” 3rd International Conference on Intelligent Sustainable Systems (ICISS), Thoothukudi, India, pp. 362-365, 2020, doi: 10.1109/ICISS49785.2020.9316090.
[4] D. Kim, M. Arsalan, K. Park, “Convolutional Neural Network-Based Shadow Detection in Images Using Visible Light Camera Sensor,” Sensors, vol. 18, no. 4, p. 960, 2018, doi: 10.3390/s18040960.
[5] F. Charli et al., “Implementasi Metode Faster Region Convolutional Neural Network (Faster R-CNN) Untuk Pengenalan Jenis Burung Lovebird,” Journal of Information Technology Ampera, vol. 1, no. 3, pp. 185-197, 2020.
[6] J. Bao et al., “ShadowDeNet: A Moving Target Shadow Detection Network for Video SAR,” Remote Sensing, vol. 14, no. 2, p. 320, 2022, doi: 10.3390/rs14020320.
[7] J. Pardede, H. Hardiansah, “Deteksi Objek Kereta Api menggunakan Metode Faster R-CNN dengan Arsitektur VGG 16,” MIND Journal, vol. 7, no. 1, pp. 21-36, 2022, doi: 10.26760/mindjournal.v7i1.21-36.
[8] L. Qu et al., “Evaluation Of Shadow Features,” IET Computer Vision, vol. 12, no. 1, p. 95-103, 2017, https://doi.org/10.1049/iet-cvi.2017.0159.
[9] Lee et al., “Moving Shadow Detection from Background Image and Deep Learning,” in: Huang, F., Sugimoto, A. (eds) Image and Video Technology – PSIVT 2015 Workshops, PSIVT 2015, Lecture Notes in Computer Science, vol. 9555, 2016.
[10] S. Brutzer, B. Hoferlin, G. Heidemann, “Evaluation of background subtraction techniques for video surveillance,” in CVPR, Providence, RI, pp. 1937-1944, 2011.
[11] S. Ren et al., “Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks,” in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 39, no. 6, pp. 1137-1149, 2017, doi: 10.1109/TPAMI.2016.2577031.
[12] Sarda et al., “Object Detection for Autonomous Driving using YOLO [You Only Look Once] algorithm,” in 2021 Third International Conference on Intelligent Communication Technologies and Virtual Mobile Networks (ICICV), pp. 1370-1374, 2021.
[13] T. M. Hoang et al., “Road Lane Detection Robust to Shadows Based on a Fuzzy System Using a Visible Light Camera Sensor,” Sensors, vol. 17, no. 11, p. 2475, 2017, doi: 10.3390/s17112475.
