Increasing the tactical effectiveness of precision-guided firearms in group usage

Infrared thermal imaging is a technique of visualization that captures the heat energy emitted by objects. Unlike systems dependent on visible light, thermal imaging works well in day and night conditions. This capability provides great advantages, especially under environmental constraints such as fog, rain, or snow. Due to its low susceptibility to environmental factors, thermal imaging systems are a reliable solution for critical applications in military and civilian domains. The ability to detect adversaries in heavy fog or in total darkness during military operations improves the outcome of a mission on the field. Thermographic cameras also find extensive use in civilian security screenings, fire detection, and search-and-rescue missions. Despite the wide potential of thermal imaging, its practical implementation on edge devices is fraught with challenges: there is a strong need to optimize performance and energy efficiency. With limited processing power, low energy capacity, and compact hardware, edge devices require innovative solutions, such as the development of energy-efficient algorithms and lightweight AI models. While deep learning-based object detection models improve the accuracy of thermal imaging considerably, they also involve a high degree of computational requirements, hence a delicate balance of performance and energy consumption arises on edge devices. The main contribution of this thesis is the generic framework for developing energy-efficient and high-performance solutions to object detection and tracking problems in thermal imaging on edge devices. Energy consumption, processing capacity, and object detection performance of various edge device platforms such as the NVIDIA Jetson Nano, Xavier NX, Orin Nano, Orin NX, and Rockchip RK3588 are studied. These devices were utilized for the implementation of thermal image analysis using AI-based object detection models such as YOLOX, YOLOv8, YOLOv9, YOLOv10, and Gold YOLO. The accuracy rate, frame per second, and computational cost of the models have been compared. In this study, the newly introduced RSH and RSHAY in this thesis were applied for finding a suitable AI model that can work best for the edge hardware and for selecting the most efficient system across devices. To strengthen performance in real time, there was the optimization of object detection models, with hardware accelerators utilized actively. This model came out as the best combination of edge hardware and object detection model that was efficient in carrying out thermal image analysis with the YOLOv8 PostDas model on the Jetson Orin Nano edge device at an input resolution of 512x512 pixels. The second crucial contribution provided by this paper is regarding the development of the communicating infrastructure that lets edge coordination happen. At the same time, this layer makes object detection propagate on this server for analysis purposes, wherein the contribution has gone for laying down centralized coordination, altogether new approach and thus contributed toward object tracking with facility brought to coordination amidst several associated devices. Hence, facilitating centralized analysis gave room among various devices' coordination through object tracking between numerous devices being able to avail that resource. The conclusions of this thesis contribute to the improvement of the performance of thermal imaging systems while optimizing energy efficiency. The results of the research will provide scalable solutions applicable in a wide range of systems, from battery-operated portable surveillance devices to unmanned vehicles, contributing both to academic research and industrial applications.