Camera Localization Filtering Unobservable Feature Points

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional visual SLAM technologies face challenges in camera localization due to the inclusion of feature points observable from multiple viewing angles, leading to invalid calculations and increased mismatch probabilities when the camera observes the environment from a single angle, resulting in localization failures.

Innovation Solution

A method that determines the location information and viewing-angle area information of each feature point during map building, filters out unobservable feature points during localization, and matches feature points in the target image with those in the environment map that are observable from the current viewing angle, reducing calculation and improving localization accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If feature points from multiple viewing angles are included in the environment map, then the environment map becomes more complete and comprehensive, but the calculation amount increases and mismatch probability increases when the camera observes from a single angle

Engineering Contradiction:
Improvelocalization accuracyVSAvoidcalculation complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies preliminary action by pre-calculating and storing viewing-angle area information for each feature point during the map building phase. This allows the system to quickly filter observable feature points during localization without performing complex calculations in real-time, thus resolving the contradiction between comprehensive mapping and calculation complexity

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent segments the feature point set by dividing the target environment into multiple viewing-angle areas and associating each feature point with its observable viewing-angle area. This segmentation enables the system to selectively use only relevant feature points for matching based on the current camera viewpoint, reducing calculation complexity while maintaining localization accuracy

Inventive Principle:
Principle #1Segmentation

2Productivity

If feature points observable from multiple viewing angles are used for matching, then more feature points are available for matching, but invalid calculations increase and mismatch probability increases

Engineering Contradiction:
Improvematching efficiencyVSAvoidmatching accuracy
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent applies local quality by making different feature points have different properties based on their viewing-angle areas. Each feature point is tagged with its observable viewing-angle information, allowing the system to selectively apply matching operations only to feature points that are actually observable from the current camera position. This improves both matching efficiency and accuracy by eliminating invalid matching operations

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP3813014B1Camera localization method, terminal and storage medium
Publication Date: 2023.08.23 TENCENT TECHNOLOGY (SHENZHEN) CO LTD

AI summary

Disclosed is a camera localization method, comprising: acquiring an environmental map of a target environment; predicting, according to position information of a camera when photographing a historical image before the photographing of a target image, the position of the camera when photographing the target image so as to obtain the predicted position information of the camera; filtering out feature points, which cannot be currently observed by the camera, in the environmental map according to the predicted position information of the camera, position information of each feature point in the environmental map and visual angle region information of each feature point; and matching feature points in the target image and residual feature points in the environmental map after filtering so as to obtain a feature point correlation, and determining the position information of the camera according to the feature point correlation.