Visual-Inertial Pose Estimation Amid Dynamic Object Motion

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing visual odometry methods face challenges in accurately estimating pose information, especially when dynamic objects are present, leading to reduced accuracy and increased resource consumption due to tracking losses and outlier selection issues.

Innovation Solution

A visual inertial odometry method and apparatus that combines visual and inertial data to accurately calculate pose information by distinguishing between camera motion and dynamic object motion, using a processor to select feature points, estimate rotation and translation information, and correct estimates based on inertial and visual data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If visual odometry methods are used to estimate pose information, then positioning can be achieved, but accuracy is reduced due to tracking losses and outlier selection issues when dynamic objects are present

Engineering Contradiction:
Improvepose information accuracyVSAvoidtracking performance
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent segments the feature points into two categories: those associated with dynamic objects and those associated with static objects. By separating these feature points and processing them differently, the system maintains accurate tracking of static objects while handling dynamic objects as outliers, thereby improving overall pose estimation accuracy despite the presence of moving objects in the scene.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts and removes feature points associated with dynamic objects from the tracking process. By identifying and taking out these outlier feature points that would otherwise degrade tracking performance, the system maintains reliable tracking of static objects and improves the accuracy of pose information estimation.

Inventive Principle:
Principle #2Taking out (Extraction)

2Quantity of substance

If feature points from dynamic objects are included in tracking, then more feature points are available for calculation, but resource consumption increases due to repetitively selecting and removing outliers

Engineering Contradiction:
Improvenumber of feature pointsVSAvoidresource consumption
Core Design Contradiction:
Quantity of substanceVSUse of energy by moving object

Solution Approach 1:

The patent performs preliminary classification of feature points to identify which ones are associated with dynamic objects before the main tracking and calculation process. By pre-identifying and removing these outlier feature points in advance, the system avoids the need for repetitive outlier detection and removal during processing, thereby reducing computational resource consumption while maintaining the benefit of using numerous feature points from static objects.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If homography is used to calculate depth and position, then positioning can be performed, but tracking loss occurs when tracking dynamic objects

Engineering Contradiction:
Improveposition calculation accuracyVSAvoidtracking continuity
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent segments the tracking process by distinguishing between feature points on static objects and those on dynamic objects. Homography-based position calculation is applied only to feature points on static objects, ensuring continuous and reliable tracking. Feature points on dynamic objects are identified as outliers and excluded from this calculation, preventing tracking loss while maintaining accurate position estimation for the majority of the scene.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP3633616B1Method and apparatus for outputting pose information
Publication Date: 2025.05.21 SAMSUNG ELECTRONICS CO LTD
  • EP3633616B1 patent drawingFigure 1
  • EP3633616B1 patent drawingFigure 2
  • EP3633616B1 patent drawingFigure 3

AI summary

A method and apparatus for outputting pose information selects first points from a frames captured by a first sensor, estimates rotation information between the frames based on motion information sensed by a second sensor, corrects the estimated rotation information based on third points, the third points being remaining points when second points corresponding to a dynamic object are excluded from the first points, obtains translation information between the frames based on the third points and the corrected rotation information, and outputs the corrected rotation information and the translation information.