Visual Ego-Motion Estimation for GNSS-Denied Position Tracking

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Autonomous vehicles and similar devices face challenges in accurately determining their position in obstructed regions where Global Navigation Satellite System (GNSS) signals are unavailable, relying on odometry for navigation which can lead to increased positional errors over time.

Innovation Solution

A processor-implemented method using an ego-motion model and attention model based on neural networks to estimate ego-motion information from input images, combining short-term and long-term motion data to correct and refine positional tracking, thereby enhancing navigation accuracy in GNSS-denied environments.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If odometry is used to track relative position in GNSS-obstructed regions, then the device can maintain operation without GNSS signals, but positional accuracy deteriorates over time due to accumulated errors

Engineering Contradiction:
Improveoperation continuity in GNSS-denied environmentsVSAvoidpositioning accuracy
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The system uses a feedback mechanism where the neural network continuously processes sequential image data to estimate ego-motion, and this estimation is fed back to correct the odometry-based position. The loss function compares predicted positions with actual positions (when available) to train and refine the model, creating a closed-loop system that reduces accumulated errors over time.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The neural network acts as an intermediary between the odometry system and the final position estimate. It processes visual information from image sequences to generate ego-motion estimates that serve as a corrective intermediary layer, bridging the gap between odometry data and accurate positioning without requiring direct GNSS signals.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Device complexity

If traditional odometry methods are used for position tracking, then the system remains simple to implement, but measurement precision deteriorates due to error accumulation over time

Engineering Contradiction:
Improvesystem implementation simplicityVSAvoidposition tracking accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent replaces traditional mechanical/mathematical odometry methods with a neural network-based visual estimation system. Instead of relying on wheel encoders or inertial sensors that accumulate errors, the system substitutes these with a deep learning model that processes image sequences to estimate motion, achieving higher accuracy without proportionally increasing system complexity.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system changes the parameters used for motion estimation from physical sensor data (odometry) to visual feature data from image sequences. By transforming the input parameters from mechanical measurements to visual descriptors processed by the neural network, the system achieves better precision while maintaining reasonable complexity through efficient network architecture design.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3719747B1Method and apparatus with ego motion information estimation
Publication Date: 2022.09.07 SAMSUNG ELECTRONICS CO LTD
  • EP3719747B1 patent drawingFigure 1
  • EP3719747B1 patent drawingFigure 2
  • EP3719747B1 patent drawingFigure 3

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.