Fast VSLAM Initialization Using Single Reference Image

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Traditional Visual Simultaneous Localization and Mapping (VSLAM) methods require known references, precise initial images, and user input for initialization, leading to frustrating and unnatural augmented reality experiences due to complex camera motion requirements.

Innovation Solution

The Fast VSLAM Initialization module (FVI) enables real-time, single-reference-image-based initialization and tracking without prior knowledge of the environment, using interest point detection and 6DoF camera pose estimation, allowing for immediate and intuitive augmented reality updates without additional sensors or complex systems.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional VSLAM methods are used with known references and precise initial images, then tracking accuracy is improved, but user operation complexity increases and initialization time is extended

Engineering Contradiction:
Improvetracking accuracyVSAvoiduser operation complexity
Core Design Contradiction:
Measurement precisionVSEase of operation

Solution Approach 1:

The system performs self-initialization by automatically detecting interest points and estimating camera pose from the first captured image without requiring user input to select reference images or perform specific camera motions. The algorithm autonomously initializes the 3D target and begins tracking, eliminating the need for users to understand or execute complex initialization sequences.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system performs preliminary interest point detection and camera pose estimation from the very first image captured, establishing initial 3D target parameters before any user-directed camera motions are required. This preliminary initialization allows tracking to begin immediately with subsequent images without requiring users to complete predefined motion sequences.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If traditional VSLAM initialization with two reference images is used, then 3D target initialization accuracy is improved, but initialization time and user input requirements increase

Engineering Contradiction:
Improve3D target initialization accuracyVSAvoidinitialization time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary initialization using the first captured image by detecting interest points and estimating camera pose, establishing initial 3D target parameters immediately. This preliminary action eliminates the need to wait for a second reference image with sufficient baseline, allowing tracking to begin without the traditional initialization delay.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The initialization process is segmented into independent steps: interest point detection from the first image, camera pose estimation, and 3D target parameter initialization. This segmentation allows the system to complete initialization with the first image alone, rather than requiring the coordinated capture of two specific reference images with adequate baseline separation.

Inventive Principle:
Principle #1Segmentation

3Reliability

If direct user input is required to select reference images and provide visual targets, then tracking reliability is improved, but ease of operation deteriorates

Engineering Contradiction:
Improvetracking reliabilityVSAvoiduser input requirements
Core Design Contradiction:
ReliabilityVSEase of operation

Solution Approach 1:

The system autonomously selects and processes the first captured image for initialization, automatically detecting interest points and estimating camera pose without requiring user selection of reference images or provision of visual targets. The algorithm self-manages the entire initialization process, eliminating user input requirements while maintaining tracking reliability through robust interest point detection and pose estimation.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP2915140B1Fast initialization for monocular visual slam
Publication Date: 2018.10.17 QUALCOMM INC
  • EP2915140B1 patent drawingFigure 1
  • EP2915140B1 patent drawingFigure 2
  • EP2915140B1 patent drawingFigure 3

AI summary

Apparatuses and methods for fast visual simultaneous localization and mapping are described. In one embodiment, a three-dimensional (3D) target is initialized immediately from a first reference image and prior to processing a subsequent image. In one embodiment, one or more subsequent reference images are processed, and the 3D target is tracked in six degrees of freedom. In one embodiment, the 3D target is refined based on the processed the one or more subsequent images.