Multi-Camera IMU Calibration Using 3D Pose Alignment

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional online calibration methods for cameras and IMUs in visual and inertial navigation systems result in low accuracy parameters due to the difficulty in accurately reflecting the relative position and attitude transformation relationship between the camera and IMU, which are mounted at different positions on a carrier platform.

Innovation Solution

A method involving the use of a terminal to obtain image sequences from multiple cameras, determine matching feature points, filter these points based on a relationship matrix, convert them into three-dimensionally distributed points, and optimize calibration parameters between the IMU and cameras by aligning poses, using techniques such as superglue feature point matching, deep learning algorithms, and preintegration to improve accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If online calibration is performed on a single camera and IMU using conventional methods, then the calibration process is simple, but the calibration accuracy is low

Engineering Contradiction:
Improvecalibration accuracyVSAvoidcalibration system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the calibration process into multiple independent calibration units, each corresponding to a different camera viewing direction (front, rear, left, right). Each calibration unit processes image sequences from its specific camera independently, then integrates results. This segmentation allows comprehensive multi-directional calibration while maintaining manageable complexity through modular processing.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transitions from single-camera 2D image calibration to multi-camera 3D spatial calibration by introducing the dimension of viewing direction. Multiple cameras capture images from different spatial angles, and the system integrates these multi-dimensional observations to achieve accurate 3D pose estimation and calibration parameters.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Measurement precision

If multiple cameras are used for calibration, then the calibration accuracy improves, but the computational complexity increases

Engineering Contradiction:
Improvecalibration accuracyVSAvoidcomputational load
Core Design Contradiction:
Measurement precisionVSPower

Solution Approach 1:

The patent divides the computational workload into separate calibration units, each handling one camera's image sequences independently. This segmentation parallelizes computation across multiple processing threads or devices, reducing the computational burden on any single processing unit while maintaining the benefits of multi-camera calibration.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary feature point extraction and matching within each calibration unit before integration. By pre-processing image sequences to extract key feature points and establish initial correspondences, the system reduces the computational complexity of the subsequent pose estimation and calibration parameter optimization stages.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If feature point matching is performed without filtering, then the processing speed is fast, but the calibration precision is low due to noise

Engineering Contradiction:
Improvecalibration precisionVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary filtering of feature points based on geometric constraints (epipolar geometry, fundamental matrix) before using them for calibration. This pre-filtering removes obvious outliers and noise while preserving valid feature correspondences, improving calibration precision without requiring extensive iterative refinement later.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements efficient filtering that quickly eliminates invalid feature points using geometric consistency checks, allowing the system to skip detailed verification of obviously incorrect matches. This rapid filtering approach maintains high precision by removing noise while minimizing the time spent on feature point processing.

Inventive Principle:
Principle #21Skipping (Rushing through)

Data Source

PatentEP4386676B1Method and apparatus for calibrating cameras and inertial measurement unit, and computer device
Publication Date: 2026.02.11 TENCENT TECHNOLOGY (SHENZHEN) CO LTD
  • EP4386676B1 patent drawingFigure 1
  • EP4386676B1 patent drawingFigure 2
  • EP4386676B1 patent drawingFigure 3a

AI summary

This application relates to a method and apparatus for calibrating cameras and an inertial measurement unit (IMU), a computer device, a storage medium, and a computer program product. The method may be applied to application scenarios such as online calibration of autonomous driving, electronic maps, and sensors. The method includes: obtaining image sequences acquired by cameras (S202); matching feature points of different images in the image sequences, to obtain matching feature points (S204); filtering the matching feature points to obtain target feature points (S206); converting the target feature points into three-dimensionally distributed points, and determining poses of the cameras based on the three-dimensionally distributed points and pixels of the images in the image sequences that correspond to the three-dimensionally distributed points (S208); and optimizing calibration parameter between an IMU and the cameras according to the poses of the cameras and a pose of the IMU (S210).