Bird's-Eye Motion Reconstruction From First-Person Time-Series Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques fail to reconstruct the motion of observing cameras and peripheral moving bodies, are limited to environments with stable static landmarks, and struggle with complex motions and GPS interference.
Innovation Solution
A bird's-eye data generating device that uses time-series data from two-dimensional observation information to generate bird's-eye data expressing the on-ground motion of an observing moving body and other moving bodies, even without detected static landmarks, using a trained model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If techniques like MonoLoco or CubeSLAM are used for position estimation, then self-position can be estimated from first-person images, but the motion of the observing camera and peripheral moving bodies cannot be reconstructed
Solution Approach 1:
The patent creates a virtual bird's-eye view copy of the first-person images to reconstruct both self-motion and peripheral moving body motions. By generating a bird's-eye viewpoint image from the first-person perspective images and extracting motion information from this virtual copy, the system recovers complete motion data without requiring actual bird's-eye cameras or GPS
Solution Approach 2:
The patent introduces a bird's-eye viewpoint image generation unit as an intermediary that transforms first-person images into a bird's-eye perspective representation. This intermediary virtual image serves as a bridge to extract both self-motion and peripheral body motion information, enabling complete motion reconstruction from first-person data alone
2Adaptability or versatility
If CubeSLAM is used for self-position estimation, then positioning can be achieved using moving bodies as landmarks, but the method is limited to environments with stable static landmarks and simple rigid-body motions
Solution Approach 1:
The patent employs dynamic motion models that can handle complex non-rigid motions of peripheral bodies. Instead of assuming simple rigid-body motions as in CubeSLAM, the system uses learned motion patterns from training data to accurately track and predict motions of diverse moving bodies in varying environmental conditions
Solution Approach 2:
The patent changes the fundamental parameters of the positioning approach by using bird's-eye viewpoint image generation and motion extraction from virtual images rather than traditional landmark-based methods. This parameter change enables operation in environments without stable static landmarks while maintaining reliable position reconstruction through learned motion dynamics
3Measurement precision
If GNSS is used for position estimation, then self-position can be obtained through satellite signals, but the method fails in environments with GPS interference from high-rise buildings
Solution Approach 1:
The patent replaces the mechanical/GPS-based positioning system with an optical-computer vision-based system. By substituting satellite signal reception with image-based bird's-eye viewpoint generation and motion extraction, the system eliminates dependency on GPS signals and operates reliably in GPS-denied environments like urban canyons with high-rise buildings
4Measurement precision
If JP-A No. 2021-77287 is used for position estimation, then bird's-eye position can be estimated by comparing motion characteristics, but the method cannot be applied when bird's-eye viewpoint images are not acquired
Solution Approach 1:
The patent creates a virtual copy of the bird's-eye viewpoint from first-person images, enabling motion comparison and position estimation without requiring actual bird's-eye camera data. This virtual copying approach extends the method's applicability to scenarios where only first-person images are available
Data Source
AI summary
The present disclosure provides a bird's-eye data generating device with an acquiring section acquiring time-series data; and a generating section generating bird's-eye data.