Patrol Robot Pose Correction via 3D Model Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In patrol inspection tasks, simultaneous location and mapping (SLAM) algorithms often result in inaccurate positioning due to weather and environmental changes, causing inspection robots to miss predetermined inspection points and leading to false detection, with existing solutions like improving SLAM accuracy or adding markers being resource-intensive and ineffective.
Innovation Solution
A method that determines offsets between to-be-inspected and predetermined inspection points by scanning the environment to generate simulated 3D object models, comparing them to pre-stored models to adjust the robot's pose, and capturing 2D images after adjustment, combining SLAM positioning with 3D object recognition for accurate targeting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If SLAM algorithm is used for positioning, then the inspection robot can navigate autonomously, but positioning accuracy deteriorates due to weather and environmental changes
Solution Approach 1:
The patent introduces 3D object recognition as an intermediary technology between SLAM positioning and inspection point identification. By recognizing 3D models of inspection targets, the system can accurately determine inspection point locations even when SLAM positioning deviates, thus resolving the contradiction between autonomous navigation and positioning accuracy.
Solution Approach 2:
The patent replaces reliance on SLAM's mechanical positioning system with visual-based 3D object recognition. Instead of depending solely on sensor-based SLAM positioning that deteriorates in varying environments, the system uses camera-based 3D modeling and recognition to substitute and correct positioning errors.
2Measurement precision
If SLAM positioning accuracy is improved, then inspection point location accuracy improves, but resource consumption increases
Solution Approach 1:
The patent applies partial action by using 3D object recognition only at critical moments when SLAM positioning is suspected to be inaccurate, rather than continuously improving SLAM accuracy. This selective approach achieves the needed positioning precision without the continuous resource consumption of enhanced SLAM systems.
Solution Approach 2:
The patent creates 3D model copies of inspection targets for recognition and comparison. By using these digital 3D copies, the system can accurately identify inspection points without requiring resource-intensive real-time SLAM optimization, achieving precision through model matching rather than continuous positioning improvement.
3Measurement precision
If markers are added in the inspection region, then positioning accuracy improves, but device complexity increases
Solution Approach 1:
The patent enables the inspection targets themselves to serve as positioning references through their inherent 3D features. Instead of adding external markers to the environment, the system uses the targets' own geometric characteristics for recognition and positioning, making the targets self-service their positioning function without increasing environmental complexity.
Solution Approach 2:
The patent utilizes 3D geometric features and visual characteristics of inspection targets for recognition, analogous to using visual cues instead of physical markers. By recognizing shape, size, and spatial characteristics of targets, the system achieves positioning accuracy through visual differentiation rather than physical marker addition.
Data Source
AI summary
A patrol inspection method includes: determining that an offset exists between a position of a to-be-inspected point and a position of a predetermined inspection point; obtaining a simulated three-dimensional (3D) object model by scanning a surrounding environment of a robot at the to-be-inspected point; comparing the simulated 3D object model and a pre-stored 3D object model to obtain adjustment information, the pre-stored 3D object model being 3D object information obtained by scanning the surrounding environment of the robot at the predetermined inspection point; based on the adjustment information, adjusting a pose of the robot; and capturing a two-dimensional (2D) image of an inspection target by the robot after adjustment.