Description:The present invention relates to an AI-powered cognitive system designed for real-time autonomous drone surveillance, incorporating advanced deep learning techniques, edge AI computing, and optimized video streaming protocols. The system utilizes the YOLOv8-Seg model, a state-of-the-art object detection and segmentation algorithm, integrated with GPU-accelerated processing on an NVIDIA Jetson Orin Nano. This combination enables low-latency, high-accuracy face detection and recognition in dynamic environments.
The drone system is built on an F450 quadcopter frame, ensuring stability and scalability for various surveillance applications. The onboard computational architecture comprises a dual-system setup with a Raspberry Pi 3 Model B+ managing telemetry and initial data capture, while the Jetson Orin Nano performs computationally intensive AI inference tasks, ensuring real-time performance. The camera module, a MIPI CSI-based high-resolution sensor, captures video data, which is processed through GStreamer for efficient encoding and transmission. The ground station, also equipped with an NVIDIA Jetson Orin Nano, receives the video feed, decodes it, and applies AI-driven face recognition algorithms to identify individuals with high precision.
The communication framework integrates Wi-Fi for video streaming and the MAVLink protocol for telemetry exchange, allowing seamless real-time feedback between the drone and the ground station. This ensures stable command transmission with minimal latency (~40ms), while video data maintains an average latency of ~120ms with ~0.5% packet loss, enabling uninterrupted surveillance operations.
The system's AI model is fine-tuned on a custom Student Face Dataset, enhancing detection accuracy for real-world applications. The model leverages advanced loss functions, such as defocus loss, to improve performance under challenging conditions, including low-light environments, motion blur, and occlusions. The power management system, optimized for energy efficiency, employs a 3S 3000mAh LiPo battery, extending flight endurance while maintaining stable computational performance.
The drone is capable of real-time autonomous decision-making, adapting to varying environmental conditions by dynamically adjusting camera focus, flight altitude, and movement patterns. The hardware-software synergy ensures smooth integration of real-time object detection, video streaming, and telemetry analysis. Experimental evaluations demonstrate the system's superior performance, achieving 95.39% mAP@0.5 accuracy with a 98% confidence score for unknown face detection.
By leveraging CUDA 12.6 and cuDNN 9.3 for AI inference, the system achieves frame rates of up to 50 FPS, significantly outperforming conventional cloud-based surveillance models. The modularity of the architecture allows for easy scalability, supporting additional AI models, sensors, or upgraded hardware, making it a versatile solution for security, reconnaissance, and real-time monitoring applications.
, Claims:1. A drone-based surveillance system that integrates a deep learning-based face detection model with GPU-accelerated inference for real-time monitoring.
2. A modular AI architecture combining Raspberry Pi 3 Model B+ and NVIDIA Jetson Orin Nano for workload distribution and optimal processing efficiency.
3. An optimized YOLOv8-Seg model fine-tuned on a custom dataset for high-accuracy face recognition.
4. A real-time video streaming system utilizing GStreamer for low-latency data transmission.
5. An energy-efficient power management system ensuring prolonged flight endurance.
6. A telemetry exchange system based on MAVLink, providing stable communication between the drone and ground station.
7. A GPU-accelerated inference pipeline leveraging CUDA and cuDNN for optimized object detection performance.
8. An adaptable and scalable framework supporting integration with additional sensors and AI models.