Autonomous on-board object and phenomenon detection system

¹ Siberian Federal University, 79 Svobodny pr., Krasnoyarsk, 660041, Russian Federation
² Reshetnev Siberian State University of Science and Technology, 31, Krasnoyarsky Rabochy Ave, Krasnoyarsk, 660037, Russian Federation

^* Corresponding author: oleslav@mail.ru

Abstract

This paper presents the design, implementation, and evaluation of an autonomous on-board object and phenomenon detection system optimized for real-time performance and resource-constrained environments. The proposed framework integrates a multimodal sensor array, including RGB cameras and LiDAR, with lightweight deep learning algorithms for object detection, tracking, and classification. Four state-of-the-art detection models - YOLO, DETR, CenterNet, and M2Det - were examined using the Lacmus Drone Dataset, a publicly available collection of over 3,000 aerial images. Experimental results highlight that no single model consistently outperforms the others: YOLO excels in real-time scenarios due to its fast inference speed, whereas DETR achieves the highest accuracy at the expense of greater computational complexity. CenterNet offers a balanced approach for detecting smaller objects, and M2Det demonstrates strong performance in densely populated urban scenes. Overall, these findings emphasize the importance of selecting model architectures based on mission requirements and hardware constraints, paving the way for more efficient and adaptive autonomous detection systems.