UAV Relative Pose Estimation Using Trajectory-Based Visual Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for calculating the relative pose between unmanned aerial, naval, and ground vehicles face challenges such as extra weight and power consumption due to visual markers, limited visibility, perceptual aliasing, and performance degradation in varying weather conditions, especially for small-sized UAVs and long distances.
Innovation Solution
A camera-based and direct observation method using a deep learning model to calculate the relative position between UAVs by having one UAV perform a predefined movement, recording its trajectory, training a deep learning model with ground truth data, and using bounding box information for tracking and feature extraction to determine the 6-DOF relative pose.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If visual markers are placed on UAVs to calculate relative pose, then relative position determination is enabled, but extra weight is added and power consumption increases
Solution Approach 1:
The patent extracts and removes the visual markers from the system. Instead of placing physical markers on UAVs, the system uses the UAVs themselves as observation targets. The relative pose is calculated by one UAV observing another UAV's movement and trajectory, eliminating the need for additional marker components and their associated weight and power consumption.
2Measurement precision
If visual markers are placed on UAVs, then relative pose can be calculated, but visibility is limited and the method does not work at long distances
Solution Approach 1:
The patent uses image copying and trajectory analysis instead of direct visual marker recognition. By capturing sequential images of the target UAV's movement and analyzing the trajectory pattern (such as eight-shaped trajectories), the system can calculate relative pose at long distances where the UAV appears as a moving point rather than requiring detailed visual marker recognition.
3Device complexity
If a single UAV is observed as a point, then detection is simplified, but only one degree of freedom is limited and other 5 degrees of freedom cannot be predicted
Solution Approach 1:
The patent transforms the static point observation into dynamic trajectory analysis. Instead of trying to detect all 6 degrees of freedom simultaneously from a single point observation, the system analyzes the dynamic movement pattern of the UAV over time. By having the target UAV perform predefined movements (such as eight-shaped trajectories) and observing the resulting trajectory patterns, the system can infer complete 6-DOF pose information from simplified point observations.
4Measurement precision
If indirect observation methods are used with common area observation, then relative pose can be estimated, but perceptual aliasing errors occur when robots think they are in the same region
Solution Approach 1:
The patent segments the observation process into distinct phases: one UAV performs a predefined movement pattern while the other observes. This active-passive segmentation eliminates the perceptual aliasing problem because the moving UAV creates a unique, time-varying trajectory pattern that cannot be confused with static environmental features or other UAV positions. The trajectory itself becomes the unique identifier rather than relying on ambiguous area-based recognition.
Data Source
AI summary
A camera-based and direct observation based relative position determination method for multiple unmanned aerial, naval and ground vehicles is provided. The method calculates the relative position between the relevant vehicles in multiple UAV, UNV and UGV systems.