Estimation device

The estimation device enhances camera pose estimation accuracy by using a model to generate images from any camera pose, addressing low-precision position issues in drive recorders, enabling precise facility identification in facility inspections.

WO2026126304A1PCT designated stage Publication Date: 2026-06-18NT T INC
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
NT T INC
Filing Date
2024-12-09
Publication Date
2026-06-18

AI Technical Summary

⚠Technical Problem

Existing image capture systems, such as drive recorders, struggle with low-precision position information, making it difficult to accurately identify facilities and associate them with management numbers, especially in environments with high-rise buildings, leading to challenges in facility inspections.

⚗Method used

An estimation device that uses a model capable of generating images from any camera pose, such as NeRF, to estimate the camera orientation by extracting candidate poses from a database and performing iterative calculations with feature matching, ensuring accurate camera pose estimation.

🎯Benefits of technology

Improves the accuracy of camera pose estimation by using candidate poses with similar global features and iterative calculations, reducing the risk of local minima and ensuring precise facility identification in facility inspections.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2024043466_18062026_PF_FP_ABST
    Figure JP2024043466_18062026_PF_FP_ABST
Patent Text Reader

Abstract

An estimation device (10) according to the present disclosure comprises: an extraction unit (11) that extracts candidate camera orientations C1-Cn; a determination unit (12) that performs feature matching of a query image and a generated image generated by inputting each of the candidate camera orientations C1-Cn into a model, and determines the camera orientation of the generated image having the highest matching rate with the query image as an initial value of the camera orientation; and an estimation unit (13) that estimates the camera orientation of the query image by repeated calculations using the determined camera orientation as the initial value. The extraction unit (11) extracts the camera orientation of search images having the closest global features with the query image as the candidate camera orientations C1-Cn in a region S1 in which the distance from the imaging position of the query image is no more than R1 when n = 1, and in a region Sn in which the distance from the imaging position of the search image of each of the candidate camera orientations C1-Cn-1 is no more than R0 and the distance from the imaging position of the query image is greater than Rn (Rn>=Rn-1) when n>=2.
Need to check novelty before this filing date? Find Prior Art