AI Edge Drone Tracking With Low-Bandwidth Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing drone systems face challenges in miniaturization and energy consumption due to high-resolution image signal processing and AI computing, leading to difficulties in precise object recognition and real-time object tracking, with conventional methods suffering from errors in position and posture estimation, data latency, and limited bandwidth issues.
Innovation Solution
An autonomous flight system utilizing artificial intelligence-based edge computing, which includes a mission apparatus that processes images and metainformation to detect objects, synchronizes coordinates with time, and transmits low-capacity data using SRT protocol, enabling real-time object tracking and reducing transmission delays.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution image signal processing and AI computing are performed on board the drone, then object recognition precision and real-time tracking capability are improved, but energy consumption and device weight increase significantly
Solution Approach 1:
The system divides image processing tasks between the drone (edge computing for real-time object detection) and ground control equipment (detailed analysis and tracking). The drone performs initial object detection and extracts key features, while ground equipment handles comprehensive image analysis, reducing on-board computational burden and energy consumption.
Solution Approach 2:
The patent introduces an intermediary communication system that transmits only essential data (object coordinates, detection results) from the drone to ground control equipment, rather than transmitting full-resolution images. This reduces transmission energy and allows ground equipment to perform detailed processing without requiring constant high-power on-board processing.
2Measurement precision
If high-resolution images are transmitted in real-time, then object tracking accuracy is improved, but transmission bandwidth requirements and data latency increase
Solution Approach 1:
The system extracts only the essential information needed for tracking (object coordinates, detection confidence, basic features) from the full-resolution images captured by the drone. This extracted data is transmitted to ground control equipment, which then accesses the original high-resolution images only when needed for detailed analysis, reducing transmission latency while maintaining tracking accuracy.
Solution Approach 2:
The drone performs preliminary object detection and coordinate extraction before transmission. By pre-processing images on-board to identify objects and their positions, the system prepares data in advance for efficient transmission, reducing the time required for both transmission and ground-based processing.
3Weight of moving object
If drone weight is reduced for miniaturization, then mobility and deployment ease are improved, but capacity for onboard processing and power supply are reduced
Solution Approach 1:
The system segments processing functions between lightweight on-board edge computing components (for basic object detection and coordinate extraction) and more powerful ground-based processing equipment. This allows the drone to remain lightweight while still performing essential real-time detection tasks, with complex processing handled on the ground.
Solution Approach 2:
The drone is equipped with minimal onboard AI capabilities to perform self-service object detection and coordinate extraction without requiring constant ground control intervention. This basic autonomy allows the lightweight drone to independently identify targets and transmit only relevant data, maintaining processing effectiveness despite reduced onboard capacity.
Data Source
AI summary
Disclosed is an autonomous flight system using artificial intelligence-based edge computing, the autonomous flight system including a mission apparatus configured to perform both image processing and object detection to generate metainformation including time coordinates of an object, a drone configured to apply a predetermined format and specifications to video including the metainformation to generate low-capacity data, and a ground controller configured to restore the low-capacity data to recognize a pre-learned object utilizing the metainformation, when a specific object is designated, to enhance the resolution of an image including the specific object, and to provide the image to the drone for object tracking.