Deteksi Objek Pada Framework Yolov5 Dengan Penanganan Kesilauan Cahaya Menggunakan Gabungan Arsitektur U-Net Dan Inpaint

Penulis

  • Firman Afrialdy Universitas Brawijaya, Malang
  • Rizal Setya Perdana Universitas Brawijaya, Malang
  • Candra Dewi Universitas Brawijaya, Malang

DOI:

https://doi.org/10.25126/jtiik.2025128866

Kata Kunci:

object detection, object segmentation, deep learning, yolov5, inpaint

Abstrak

CCTV telah diterapkan untuk memantau berbagai aktivitas di lingkungan Universitas Brawijaya, termasuk lalu lintas kendaraan di gerbang kampus. Pengawasan pada malam hari dalam kondisi intensitas cahaya yang rendah  merupakan tantangan tersendiri dalam penggunaan CCTV. Hal ini dikarenakan kualitas gambar yang rendah sehingga menghambat kemampuan sistem untuk mendeteksi dan mengidentifikasi objek dengan tepat. Salah satu permasalahan yang timbul dalam kasus kurangnya pencahayaan adalah munculnya flare atau kesilauan yang disebabkan oleh lampu kendaraan yang mengarah langsung ke CCTV. Oleh karena itu, pada penelitian ini digunakan segmentasi U-Net dan restorasi inpaint untuk preproses data sebelum dilakukan deteksi objek menggunakan framework YOLOv5. Hasil pengujian deteksi objek diperoleh nilai precision 0.942, recall 0.873, dan F1-Score 0.88 pada model yang dipreproses menggunakan segmentasi U-Net dan restorasi inpaint. Nilai tersebut lebih tinggi sebesar 0.032 pada precision, 0.018 pada recall, dan 0.3 pada F1-Score jika dibandingkan dengan model yang tanpa preproses.­­

 

Abstract

CCTV has been implemented to monitor various activities within Brawijaya University, including vehicle traffic at the campus gate. Surveillance at night in low light intensity conditions is a challenge in the use of CCTV. This is due to the low image quality that hampers the system's ability to detect and identify objects correctly. One of the problems that arise in the case of lack of lighting is the appearance of flares or glare caused by vehicle lights that point directly to the CCTV. Therefore, in this research, U-Net segmentation and inpaint restoration are used to preprocess data before object detection using the YOLOv5 framework. The results of object detection testing obtained precision values of 0.942, recall 0.873, and F1-Score 0.88 on models preprocessed using U-Net segmentation and inpaint restoration. These values are higher by 0.032 in precision, 0.018 in recall, and 0.3 in F1-Score when compared to the model without preprocessing.

 

Downloads

Download data is not yet available.

Referensi

BAZZI, A., MURTHY, J.S., M, S.G., LAI, W., D, P.B., PATIL, S.N. and L, H.K., 2022. ObjectDetect: A RealTime Object Detection Framework for Advanced Driver Assistant Systems Using YOLOv5. Wireless Communications and Mobile Computing, [online] 2022, p.9444360. https://doi.org/10.1155/2022/9444360.

CHEN, Y.L., WU, B.F., HUANG, H.Y. and FAN, C.J., 2011. A RealTime Vision System for Nighttime Vehicle Detection and Traffic Surveillance. IEEE Transactions on Industrial Electronics, 58(5), pp.2030–2044. https://doi.org/10.1109/TIE.2010.2055771.

GUO, F. and XU, Y.,. Vehicle Analysis System Based on DeepSORT and YOLOv5. In: 2022 3rd International Conference on Computer Vision, Image and Deep Learning & International Conference on Computer Engineering and Applications (CVIDL & ICCEA). pp.175–179. https://doi.org/10.1109/CVIDLICCEA56201.2022.9824363.

HUANG, S., HE, Y. and CHEN, X., 2021. MYOLO: A Nighttime Vehicle Detection Method Combining Mobilenet v2 and YOLO v3. Journal of Physics: Conference Series, [online] 1883(1), p.012094. https://doi.org/10.1088/17426596/1883/1/012094.

JOCHER, G., CHAURASIA, A., STOKEN, A., BOROVEC, J., NANOCODE012, KWON, Y., MICHAEL, K., TAOXIE, FANG, J., IMYHXY, LORNA, 曾逸夫(ZENG YIFU, WONG, C., ABHIRAM V, MONTES, D., WANG, Z., FATI, C., NADAR, J., LAUGHING and UNGLVKITDE, 2022. ultralytics/yolov5: v7.0 - YOLOv5 SOTA Realtime Instance Segmentation. https://doi.org/10.5281/zenodo.7347926.

KUTLIMURATOV, A., KHAMZAEV, J., KUCHKOROV, T., ANWAR, M.S. and CHOI, A., 2023. Applying Enhanced RealTime Monitoring and Counting Method for Effective Traffic Management in Tashkent. Sensors, 23(11). https://doi.org/10.3390/s23115007.

MAHAUR, B. and MISHRA, K.K., 2023. Smallobject detection based on YOLOv5 in autonomous driving systems. Pattern Recognition Letters, [online] 168, pp.115–122. https://doi.org/10.1016/j.patrec.2023.03.009.

MIAO, Y., LIU, F., HOU, T., LIU, L. and LIU, Y., 2020. A nighttime vehicle detection method based on YOLO v3. pp.6617–6621. https://doi.org/10.1109/CAC51589.2020.9326819.

PANGESTU, I., 2022. Mengenal Pengertian CCTV: Fungsi, Jenis dan Cara Kerjanya, Jasa Pembuatan Website - Metafora Indonesia Tehnology. [online] idmetafora.com. Available at: <https://idmetafora.com/news/read/1411/Mengenal-Pengertian-CCTV-Fungsi-Jenis-dan-Cara-Kerjanya.html>.

PARVIN, S., ISLAM, M.E. and ROZARIO, L.J., 2022. Nighttime Vehicle Detection Methods Based on Headlight Feature: A Review. IAENG International Journal of Computer Science, 49(1).

RONNEBERGER, O., FISCHER, P. and BROX, T., 2015. U-net: Convolutional networks for biomedical image segmentation. CoRR, [online] abs/1505.04597. Available at: <http://arxiv.org/abs/1505.04597>.

TELEA, A., 2004. An Image Inpainting Technique Based on the Fast Marching Method. Journal of Graphics Tools, 9. https://doi.org/10.1080/10867651.2004.10487596.

WANG, J., YANG, P., LIU, Y., SHANG, D., HUI, X., SONG, J. and CHEN, X., 2023. Research on Improved YOLOv5 for LowLight Environment Object Detection. Electronics, 12(14). https://doi.org/10.3390/electronics12143089.

WU, S., GE, F. and ZHANG, Y.,. A Vehicle LinePressing Detection Approach Based on YOLOv5 and DeepSort. In: 2022 IEEE 22nd International Conference on Communication Technology (ICCT). pp.1745–1749. https://doi.org/10.1109/ICCT56141.2022.10072680.

YANG, Y., CHENG, Z., YU, H., ZHANG, Y., CHENG, X., ZHANG, Z. and XIE, G., 2022. MSENet: generative image inpainting with multiscale encoder. The Visual Computer, [online] 38(8), pp.2647–2659. https://doi.org/10.1007/s00371021021430.

Diterbitkan

30-06-2025

Terbitan

Bagian

Ilmu Komputer

Cara Mengutip

Deteksi Objek Pada Framework Yolov5 Dengan Penanganan Kesilauan Cahaya Menggunakan Gabungan Arsitektur U-Net Dan Inpaint. (2025). Jurnal Teknologi Informasi Dan Ilmu Komputer, 12(3), 601-608. https://doi.org/10.25126/jtiik.2025128866