PENERAPAN CNN-LSTM DALAM SENSOR FUSION RADAR DAN LIDAR UNTUK DETEKSI OBJEK

Faiza Latifah, - and Liptia Venica, - and Muhammad Rizalul Wahid, - (2025) PENERAPAN CNN-LSTM DALAM SENSOR FUSION RADAR DAN LIDAR UNTUK DETEKSI OBJEK. S1 thesis, Universitas Pendidikan Indonesia.

Abstract

Keselamatan dan efisiensi pada sistem otonom membutuhkan kemampuan deteksi objek yang akurat. Namun, penggunaan sensor tunggal seperti radar atau lidar memiliki keterbatasan, di mana radar unggul dalam cuaca buruk tetapi memiliki resolusi spasial rendah, sementara lidar memiliki resolusi tinggi tetapi rentan terhadap cuaca buruk dan lebih mahal. Penelitian ini bertujuan untuk meningkatkan akurasi dan keandalan deteksi objek dengan mengintegrasikan data dari kedua sensor menggunakan metode sensor fusion. Penelitian ini menggunakan pendekatan research and development dengan metodologi Design Science Research Methodology (DSRM). Data dikumpulkan dari radar Continental ARS408-21 dan lidar Hokuyo UST-10LX menggunakan Robot Operating System (ROS). Metode early fusion diterapkan untuk menggabungkan data mentah sensor. Model deep learning gabungan CNN-LSTM dikembangkan untuk mengekstrak fitur spasial dan temporal dari data. Kinerja model dievaluasi menggunakan metrik akurasi, MAE, RMSE presisi, recall, dan F1-score. Hasil pengujian menunjukkan sensor fusion meningkatkan kemampuan deteksi secara signifikan. Kombinasi informasi kecepatan dari radar dan jarak yang presisi dari lidar menghasilkan identifikasi objek yang lebih akurat. Model yang dilatih berhasil mencapai akurasi pelatihan 94% dan akurasi validasi 94%, menunjukkan kemampuan pembelajaran yang efektif dan minim overfitting. Model ini mampu mendeteksi objek dengan baik bahkan saat sebagian tertutup. Keterbatasan masing-masing sensor berhasil diatasi melalui sensor fusion. Radar yang cenderung salah pada objek statis dan lidar yang tidak langsung mendeteksi kecepatan dapat saling melengkapi. Penerapan CNN-LSTM terbukti efektif dalam memproses data multimodal dan temporal, menghasilkan sistem deteksi yang akurat dan stabil. ----- Safety and efficiency in autonomous systems require accurate object detection capabilities. However, the use of a single sensor such as radar or LiDAR has limitations. Radar performs well in adverse weather but has low spatial resolution, while LiDAR offers high resolution but is vulnerable to poor weather conditions and is more expensive. This research aims to improve the accuracy and reliability of object detection by integrating data from both sensors using a sensor fusion approach. The study employs a research and development approach with the Design Science Research Methodology (DSRM). Data were collected from the Continental ARS408-21 radar and the Hokuyo UST-10LX LiDAR using the Robot Operating System (ROS). An early fusion method was applied to combine the raw data from the sensors. A hybrid CNN-LSTM deep learning model was developed to extract spatial and temporal features from the data. The model’s performance was evaluated using accuracy, MAE, RMSE, precision, recall, and F1-score metrics. The experimental results show that sensor fusion significantly improves detection capability. The combination of speed information from radar and precise distance measurements from LiDAR results in more accurate object identification. The trained model achieved 94% training accuracy and 94% validation accuracy, demonstrating effective learning capabilities with minimal overfitting. The model can detect objects effectively even when partially occluded. The limitations of each sensor were successfully addressed through sensor fusion. Radar’s tendency to misclassify static objects and LiDAR’s inability to directly detect velocity complement each other. The application of CNN-LSTM has proven effective in processing multimodal and temporal data, resulting in an accurate and stable object detection system.

	Text S_MKB_2103319_Title.pdf Download (568kB)
	Text S_MKB_2103319_Chapter1.pdf Download (317kB)
	Text S_MKB_2103319_Chapter2.pdf Restricted to Staf Perpustakaan Download (904kB)
	Text S_MKB_2103319_Chapter3.pdf Download (721kB)
	Text S_MKB_2103319_Chapter4.pdf Restricted to Staf Perpustakaan Download (1MB)
	Text S_MKB_2103319_Chapter5.pdf Download (279kB)
	Text S_MKB_2103319_Appendix.pdf Restricted to Staf Perpustakaan Download (2MB)

Item Type:	Thesis (S1)
Additional Information:	https://scholar.google.com/citations?view_op=list_works&hl=id&user=fY2b8FgAAAAJ ID SINTA Dosen Pembimbing: Liptia Venica: 6779029 Muhammad Rizalul Wahid: 6780434
Uncontrolled Keywords:	Deteksi objek, sensor fusion, radar, lidar, sistem otonom, deep learning, dan CNN-LSTM. Object detection, sensor fusion, radar, lidar, autonomous systems, deep learning, CNN-LSTM.
Subjects:	T Technology > T Technology (General)
Divisions:	UPI Kampus Purwakarta > S1 Mekatronika dan Kecerdasan Buatan
Depositing User:	Faiza Latifah
Date Deposited:	12 Sep 2025 08:56
Last Modified:	12 Sep 2025 09:00
URI:	http://repository.upi.edu/id/eprint/138410

Actions (login required)

View Item