Vision-Based UAV Stabilization Using Deep Gated Recurrent Q-Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current GPS-based positioning systems for drones are inadequate in urban and indoor environments due to insufficient satellite coverage and inaccurate sensory measurements for lateral stabilization, leading to potential crashes and hardware loss.
Innovation Solution
A vision-based self-stabilization system using deep gated recurrent Q-networks (DGRQNs) that processes raw images from a UAV's camera to estimate camera poses and predict actions to counteract lateral disturbances, enabling the drone to maintain stability without relying on GPS or conventional sensory systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If GPS-based positioning system is used for UAV positioning, then positioning can be obtained in open environments, but positioning accuracy deteriorates in urban and indoor environments due to insufficient satellite coverage
Solution Approach 1:
The patent introduces visual features from the environment as an intermediary to establish a relationship between the UAV and its surroundings. By extracting and matching visual features across multiple frames, the system creates a reference framework that enables positioning without relying on satellite signals, thus resolving the contradiction between reliability and precision in GPS-denied environments.
Solution Approach 2:
The patent replaces the GPS satellite-based positioning system with a vision-based positioning system. Instead of relying on electromagnetic signals from satellites, the system uses image processing and visual feature matching to determine UAV position and orientation, achieving accurate positioning in urban and indoor environments where GPS is unavailable.
2Ease of operation
If conventional sensory systems are used for lateral stabilization control, then state variables can be measured, but measurement accuracy remains below requirements for indoor environment applications
Solution Approach 1:
The patent uses visual features extracted from camera images as an intermediary to measure lateral displacement and orientation. Instead of relying on conventional sensors with insufficient accuracy, the system derives position and attitude information from visual feature matching, achieving the required measurement precision for indoor stabilization control.
Solution Approach 2:
The patent implements a feedback mechanism where visual features are continuously tracked across video frames, and the measured deviations from the reference position are used to generate correction commands. This closed-loop feedback system enables precise lateral stabilization by constantly comparing current visual measurements with the reference state and adjusting control inputs accordingly.
3Measurement precision
If vision-based self-stabilization system is implemented, then navigation accuracy improves in indoor environments, but system complexity increases due to deep learning algorithms
Solution Approach 1:
The patent implements a self-service mechanism where the system automatically selects and extracts visual features, matches them across frames, and generates control commands without human intervention. The deep learning model performs self-supervised learning from visual data, enabling the complex navigation task to be handled autonomously by the system itself, thus managing complexity through automation.
Solution Approach 2:
The patent creates a multi-functional vision-based control system that simultaneously performs positioning, navigation, and stabilization tasks. The same visual feature extraction and matching framework is used for multiple control objectives, reducing overall system complexity by consolidating multiple functions into a unified vision-based architecture rather than requiring separate specialized systems for each function.
Data Source
AI summary
A system and a method for vision-based self-stabilization by deep gated recurrent Q-networks (DGRQNs) for unmanned arial vehicles (UAVs) are provided. The method comprises receiving a plurality of raw images captured by a camera installed on a UAV; receiving an initial reference image for stabilization and obtaining an initial camera pose from the initial reference image; extracting a fundamental matrix between consecutive images and estimating a current camera pose relative to the initial camera pose, wherein the camera pose includes an orientation and a location of the camera; based on the estimated current camera pose, predicting an action to counteract a lateral disturbance of the UAV based on the DGRQNs; and based on the predicted action to counteract the lateral disturbance of the UAV, driving the UAV back to the initial camera pose.