Ego-Motion Estimation Using Short- and Long-Term Frame Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous devices such as vehicles and drones face challenges in accurately determining their position in obstructed areas where GNSS signals are unavailable, relying on odometry for relative position tracking which can lead to increased errors over time.
Innovation Solution
A processor-implemented method that estimates short-term and long-term ego-motion information from frame images using ego-motion models, corrects long-term ego-motion using attention information, and determines final ego-motion to track the device's path accurately, even in areas with limited or no GNSS signal availability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If odometry is used to track relative position in GNSS-obstructed regions, then the device can determine its position without GNSS signals, but measurement precision deteriorates over time due to accumulated errors
Solution Approach 1:
The patent segments the ego-motion estimation into short-term and long-term components. Short-term ego-motion is estimated from consecutive frame images using optical flow, while long-term ego-motion is estimated from keyframe images. This segmentation allows the system to leverage the accuracy of short-term optical flow for immediate position updates while using long-term keyframe matching to correct accumulated drift, thereby resolving the contradiction between maintaining position determination reliability without GNSS and preventing measurement precision degradation over time.
Solution Approach 2:
The patent implements a feedback mechanism where long-term ego-motion information is used to correct short-term ego-motion accumulation errors. The system continuously compares the accumulated short-term position estimates with long-term keyframe-based position estimates and applies corrections to eliminate drift. This feedback loop ensures that position measurement precision is maintained over extended periods while preserving the reliability of operation in GNSS-denied environments.
2Speed
If only short-term ego-motion information is used for position tracking, then the tracking is responsive to immediate movements, but measurement precision deteriorates due to error accumulation over time
Solution Approach 1:
The patent segments the temporal scope of ego-motion estimation into short-term (consecutive frames) and long-term (keyframes). The short-term component provides responsive tracking of immediate movements, while the long-term component corrects accumulated errors. This dual-timescale segmentation enables the system to maintain both responsiveness to current motion and accuracy over extended periods.
Solution Approach 2:
The patent performs preliminary estimation of short-term ego-motion from consecutive frames to provide immediate position tracking responsiveness. This short-term estimate is then refined using long-term keyframe-based ego-motion information that corrects for drift. The preliminary short-term action ensures responsive tracking while the subsequent long-term correction maintains measurement precision.
3Measurement precision
If only long-term ego-motion information is used for position tracking, then error accumulation is reduced, but responsiveness to immediate movements is reduced
Solution Approach 1:
The patent segments the ego-motion estimation process into two parallel tracks: short-term estimation from consecutive frames for immediate responsiveness, and long-term estimation from keyframes for accuracy. Both segments are integrated to produce the final position estimate, ensuring that neither responsiveness nor precision is compromised.
Solution Approach 2:
The patent merges short-term and long-term ego-motion information to achieve both responsiveness and precision. The short-term ego-motion provides immediate tracking of movements, while the long-term ego-motion corrects drift. By combining these two sources of information through fusion algorithms, the system achieves responsive tracking without sacrificing measurement precision over time.
4Measurement precision
If both short-term and long-term ego-motion models are used, then position tracking accuracy is improved, but device complexity increases
Solution Approach 1:
The patent segments the computational workload into two distinct models: a short-term ego-motion model processing consecutive frames and a long-term ego-motion model processing keyframes. This segmentation allows each model to be optimized for its specific temporal scope, improving overall measurement precision while managing complexity by distributing computational tasks rather than concentrating them in a single complex model.
Solution Approach 2:
The patent applies partial processing by using only keyframes (a subset of all frames) for long-term ego-motion estimation, rather than processing every frame at high computational cost. This partial action approach maintains measurement precision through long-term constraints while reducing overall device complexity and computational burden compared to full-frame processing.
Data Source
AI summary
A processor-implemented method includes: estimating, from frame images of consecutive frames acquired from one or more sensors, short-term ego-motion information of the one or more sensors; estimating long-term ego-motion information of the one or more sensors from the frame images; determining attention information from the short-term ego-motion information and the long-term ego-motion information; and determining final long-term ego-motion information of a current frame, of the consecutive frames, based on the long-term ego-motion information and the attention information.