DETEKSI OBJEK WAKTU NYATA PADA JETSON ORIN NANO UNTUK PEMANTAUAN KAPASITAS ORANG
DOI:
https://doi.org/10.54314/8r5f4468Keywords:
edge AI, Jetson Orin Nano, real-time object detection, confidence threshold person countingAbstract
Abstract: Real-time object detection on edge devices forces a trade-off between image quality and inference speed, yet both are often reported as a single performance figure. This study examines that trade-off on an NVIDIA Jetson Orin Nano paired with a Logitech C922 USB camera for person-capacity monitoring. The first stage tested two resolutions, three lighting conditions, and three distances to select a camera configuration. The second stage built a detectNet program with TensorRT acceleration that counts persons and bottles, raises a capacity-exceeded status when three or more persons appear, and logs every frame to CSV across six test scenarios. A resolution of 1280 x 720, normal lighting, and a 1 m distance gave the best visual quality (4 of 5), while low light and backlight reduced it to 2. At 640 x 480 the program reached 137-176 FPS with 5.7-7.3 ms latency, roughly five times the camera limit of 30 fps, which shows that inference rate and capture rate must be reported separately. Raising the confidence threshold from 0.50 to 0.70 removed low-confidence detections of 51-63%. Timing at 1280 x 720 failed (FPS reported as Inf) and temperature read 0.0 C, so these findings remain preliminary until wall-clock timing and sensor readings are corrected.
Keywords: edge AI, Jetson Orin Nano, real-time object detection, confidence threshold, person counting.
Abstrak: Deteksi objek waktu nyata di perangkat edge selalu menukar kualitas citra dengan kecepatan inferensi, tetapi keduanya sering dilaporkan sebagai satu angka performa. Penelitian ini menguji pertukaran tersebut pada NVIDIA Jetson Orin Nano dengan kamera USB Logitech C922 untuk pemantauan kapasitas orang. Tahap pertama menguji dua resolusi, tiga kondisi cahaya, dan tiga jarak untuk memilih konfigurasi kamera. Tahap kedua membangun program berbasis detectNet dan TensorRT yang menghitung orang dan botol, memberi status KAPASITAS TERLAMPAUI bila terdapat tiga orang atau lebih, dan mencatat setiap bingkai ke CSV pada enam skenario uji. Resolusi 1280 x 720, cahaya normal, dan jarak 1 m menghasilkan kualitas visual terbaik (4 dari 5), sedangkan cahaya rendah dan backlight menurunkannya ke nilai 2. Pada 640 x 480 program mencapai 137-176 FPS dengan latency 5,7-7,3 ms, sekitar lima kali batas kamera 30 fps, sehingga laju inferensi dan laju tangkap kamera harus dilaporkan terpisah. Kenaikan threshold dari 0,50 ke 0,70 menyaring deteksi berkeyakinan rendah 51-63%. Pengukuran pada 1280 x 720 gagal (FPS terbaca Inf) dan suhu terbaca 0,0 C, sehingga temuan ini masih bersifat awal sampai pengukuran waktu dinding dan pembacaan sensor diperbaiki.
Kata kunci: edge AI, Jetson Orin Nano, deteksi objek waktu nyata, threshold confidence, penghitungan orang.
Downloads
References
Aljami, H. M., Alrowais, N. A., AlAwajy, A. M., Alhrgan, S. M., Aldwaani, R. A., Alsawadi, M. S., Saqib, N. U., Alam, S. S., & Alsubaie, R. (2026). Benchmarking YOLOv8 variants for object detection efficiency on Jetson Orin NX for edge computing applications. Computers, 15(2), 74. https://doi.org/10.3390/computers15020074
Al-Mimi, H., Al-Dahoud, A., Al-Dahoud, A., & Fezari, M. (2026). Edge AI: Embedded intelligence, a review on hardware and applications. WSEAS Transactions on Information Science and Applications, 23.
Burkus, E., Lestyán, B., Dénes-Fazakas, L., & Eigner, G. (2026). Quantized vs. full-precision YOLO models on edge devices: A performance benchmark for real-time license plate detection in smart parking systems. Sensors, 26(16), 5034. https://doi.org/10.3390/s26165034
Diaz-Santos, S., Caballero-Gil, P., & Caballero-Gil, C. (2025). Real-time passenger flow analysis in tram stations using YOLO-based computer vision and edge AI on Jetson Nano. Computers, 14(11), 476. https://doi.org/10.3390/computers14110476
Edge AI in practice: A survey and deployment framework for neural networks on embedded systems. (2025). Electronics, 14(24), 4877. https://doi.org/10.3390/electronics14244877
Franklin, D. (n.d.). Hello AI World: jetson-inference [GitHub repository]. https://github.com/dusty-nv/jetson-inference
Gomulka, K., & Wozniak, P. (2026). People counting using YOLO-based detection and clustering for a mobile robot. Sensors, 26(15), 4664. https://doi.org/10.3390/s26154664
Kim, J., Kim, G., Yoshitoshi, R., & Tokuda, K. (2025). Real-time object detection for edge computing-based agricultural automation: A case study comparing the YOLOX and YOLOv12 architectures and their performance in potato harvesting systems. Sensors, 25(15), 4586. https://doi.org/10.3390/s25154586
Krišlauks, R., Tiscenko, D., Medvedevs, V., Ormanis, J., & Judvaitis, J. (2026). Ambient temperature impact on the thermal behavior and power consumption of the NVIDIA Jetson AGX Orin in an outdoor enclosure. Electronics, 15(11), 2467. https://doi.org/10.3390/electronics15112467
Mela, J. L., & Sánchez, C. G. (2025). Yolo-based power-efficient object detection on edge devices for USVs. Journal of Real-Time Image Processing, 22, 108. https://doi.org/10.1007/s11554-025-01682-2
Otero, P. M., Tellaeche, A., & Melero, M. H. (2026). Performance analysis of the YOLO object detection algorithm in embedded systems: Generated code vs. native implementation. Computation, 14(3), 67. https://doi.org/10.3390/computation14030067
Shalash, O., Khatab, E., El-Agamy, A.,
Elmokadem, L., Abouelsaad, Y., Zaki, J., El-Sayed, M., & Said, H. (2026). Benchmarking YOLOv8-YOLOv12 for real-time object detection on single-board computers. Machine Learning and Knowledge Extraction, 8(7), 204. https://doi.org/10.3390/make8070204
Su, Y., & Lu, M. (2026). Exposure-aware training for low-light object detection without target-domain data. Journal of Imaging, 12(6), 245. https://doi.org/10.3390/jimaging12060245
Syahrudin, E., Utami, E., & Hartanto, A. D. (2025). Object detection with YOLOv8 and enhanced distance estimation using OpenCV for visually impaired accessibility. JOIV: International Journal on Informatics Visualization, 9(2), 575-584. https://doi.org/10.62527/joiv.9.2.282
Wang, X., Tang, Z., Guo, J., Meng, T.,
Wang, C., Wang, T., & Jia, W. (2025). Empowering edge intelligence: A comprehensive survey on on-device AI models. ACM Computing Surveys, 57(9), 228. https://doi.org/10.1145/3724420
Wenkel, S., Alhazmi, K., Liiv, T., Alrshoud, S., & Simon, M. (2021). Confidence score: The forgotten dimension of object detection performance evaluation. Sensors, 21(13), 4350. https://doi.org/10.3390/s21134350
Yang, P., Jiang, J., & Zhang, Y. (2026). A discriminative enhancement and selective fusion method for low-light cross-spectral object detection. Sensors, 26(9), 2684. https://doi.org/10.3390/s26092684
Yao, Z., Xu, H., Zhang, H., Tu, X., & Yin, J. (2026). Improved CenterNet-based multimodal object detection for low-light and complex environments. Sensors, 26(12), 3735. https://doi.org/10.3390/s26123735
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Fitria Priyulida, Fitria Nova Hulu, David Nababan, Natha Niel Mendrofa (Author)
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
