Unmanned Vehicle Pose Determination with Reference Image Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining the pose of an unmanned driving device using acquired images result in significant errors, leading to low accuracy in path planning and environmental perception.
Innovation Solution
A method involving obtaining environment image data, predicting a pose for matching reference image data, determining pose deviation representation information, and selecting target image data to improve pose accuracy, utilizing feature extraction networks and pre-trained global and relative pose prediction models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pose data is determined using images acquired by the unmanned driving device, then path planning and environmental perception can be performed, but the accuracy of pose determination is low due to significant errors
Solution Approach 1:
The patent applies feedback by calculating pose deviation between the predicted pose (from image matching) and the actual pose (from sensor fusion). This deviation is then used to adjust and optimize the pose determination, creating a closed-loop system that continuously improves accuracy by learning from errors.
Solution Approach 2:
The patent performs preliminary action by pre-calculating and storing pose deviation information for multiple reference images. When determining the current pose, the system retrieves and utilizes this pre-computed deviation data to quickly correct the predicted pose, avoiding real-time complex calculations and improving both speed and accuracy.
2Measurement precision
If multiple reference image data are used for pose prediction, then pose accuracy can be improved, but the computational complexity and time consumption increase
Solution Approach 1:
The patent performs preliminary action by pre-calculating pose deviation information for multiple reference images during an offline or initialization phase. This pre-computed deviation data is stored and readily available when needed, allowing the system to quickly retrieve and apply corrections without performing complex real-time calculations, thus reducing time consumption while maintaining high accuracy.
Solution Approach 2:
The patent applies partial action by selectively using only the most relevant reference images and their pre-computed pose deviation information for the current pose determination, rather than processing all available reference images. This selective approach reduces computational load and time consumption while still achieving high accuracy through targeted use of deviation data.
3Measurement precision
If pose deviation representation information is calculated for all reference image data, then the accuracy of selecting target image data improves, but the computational load increases
Solution Approach 1:
The patent applies local quality by calculating pose deviation information selectively for specific reference images that are most relevant to the current environment, rather than uniformly for all reference images. The system identifies and focuses computational resources on locally important reference images, achieving high selection accuracy while reducing overall computational power consumption.
Solution Approach 2:
The patent applies partial action by calculating pose deviation information only for a subset of reference images that are deemed most useful for the current pose determination task. This selective calculation approach reduces the overall computational load and power consumption while maintaining high accuracy in target image data selection through focused analysis of critical reference images.
Data Source
AI summary
First, environment image data acquired by an unmanned driving device is obtained, and for each piece of reference image data matching the environment image data, a predicted pose of the unmanned driving device when acquiring the environment image data is determined according to an actual pose corresponding to the reference image data; and then pose deviation representation information of the reference image data is determined according to the predicted pose and actual poses corresponding to other pieces of reference image data. Finally, target image data is selected from the reference image data, and a pose of the unmanned driving device when acquiring the environment image data is determined.