Object detection and classification

Abstract

The integration of Artificial Intelligence (AI) and Robotics technologies has become pivotal in modern society, driving innovations in smart cities and revolutionizing agricultural practices. This thesis presents a comprehensive exploration of these synergies, with a primary focus on leveraging the YOLO (You Only Look Once) architecture and convolution-based models like YOLOv5, YOLOv8, and YOLOv8-OBB (Oriented bounding box) for object detection and classification tasks.The study emphasizes YOLO-OBB’s exceptional ability to detect diagonal objects, making it a valuable asset in scenarios where precise object recognition is crucial. Extensive training of these models has been conducted, particularly targeting infrastructural faults such as cracks, potholes, and insulator faults. A key highlight is the crack severity classification based on orientation, a critical factor in risk assessment and mitigation strategies. Deploying these trained models on AI-Edge devices such as Jetson boards, integrated with connected cameras, has significantly optimized inference times, a vital aspect for real-time applications demanding swift decision-making. Moreover, to overcome geographical and accessibility challenges during inspections, the study incorporates Unmanned Aerial Vehicles (UAVs). Jetson boards housing the deployed models are mounted on these UAVs, harnessing GPS capabilities for enhanced geotechnical insights and precise localization during inspections. Expanding the scope beyond urban infrastructure, the research extends the object detection approach to agriculture, particularly in crop harvesting processes. Cameras detect objects, aiding robotic arms in efficiently picking harvested crops. This integration of AI and Robotics streamlines agricultural operations, enhancing productivity and reducing labour-intensive tasks. Furthermore, the study explores the application of object detection in space exploration, specifically in lunar surface feature detection and classification. The robustness of AI-driven object detection algorithms proves instrumental in analyzing and understanding lunar terrain, contributing valuable insights to space exploration missions. In conclusion, this thesis represents a significant advancement in the fusion of AI and Robotics technologies across diverse domains. The findings not only showcase the potential for enhancing efficiency, accuracy, and automation in smart cities and agriculture but also highlight the transformative impact of AI-driven solutions in pushing the boundaries of space exploration.