Enhance Security with Robots and Artificial Intelligence

Ensuring security in critical infrastructures, industrial facilities, and remote areas presents a persistent challenge. Traditional methods, such as fixed surveillance cameras and human patrols, suffer from inherent limitations in coverage, reaction time, and analytical capabilities. Human guards are prone to fatigue, errors, and cognitive biases, while static cameras lack mobility and adaptability. To address these constraints, this thesis presents the development of an autonomous robotic patrolling system that integrates artificial intelligence, real-time data streaming, and mobile robotics to enhance surveillance efficiency. The proposed solution is designed to be platform-agnostic, allowing the robotic base to be interchangeable without modifying the perception and processing pipeline. The system utilizes an Intel RealSense D435i depth camera for real-time environment perception, while an NVIDIA Jetson Orin Nano 8GB serves as the core computational unit, executing AI-driven inference for object detection and anomaly recognition. The entire perception pipeline is built using NVIDIA DeepStream, optimizing video processing and inference performance to achieve real-time analysis. A critical aspect of this work is the dataset used for AI training. Given the challenges of acquiring large, diverse, and high-quality datasets for patrolling scenarios, this research leveraged FLUX.1-schnell, an AI generative model, to synthetically generate training images. More than 30 different prompts were tested to ensure the dataset's variability and realism, improving the model’s generalization capability in complex environments. For communication, the system employs Apache Kafka as a scalable and reliable messaging protocol. Detected events—including frames containing one or more objects, their bounding boxes, and timestamps—are transmitted to a central server. A web application then processes and visualizes this data, providing an intuitive interface for security personnel to monitor alerts and review past detections. This approach enables real-time decision-making, reducing response time to security breaches while maintaining a structured historical record of events. By integrating deep learning, mobile robotics, and high-performance edge computing, this work demonstrates a scalable and cost-efficient security solution. The system offers continuous, automated patrolling with enhanced situational awareness, reducing human workload while ensuring fast and accurate threat detection. The combination of AI-powered perception, real-time streaming, and robotic mobility represents a significant step forward in autonomous surveillance technology.