Fisheye Facial Authentication with Spherical Convolutional Networks

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional face authentication systems require individuals to position themselves in front of a camera, necessitating multiple devices for varying heights and suffer from precision issues when converting spherical images to plane images for facial recognition, particularly with fisheye cameras.

Innovation Solution

Utilizing a fisheye camera system with a ceiling-mounted camera and a platform-mounted camera, combined with a distance-measurement sensor, to capture spherical data, and employing a Spherical Convolutional Neural Network (SCNN) for robust facial recognition, with optional conversion to a plane image using Defish Eye correction for enhanced precision.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional face authentication systems use a standard camera positioned at a fixed height, then the device structure is simple, but individuals of varying heights cannot be authenticated accurately without adjusting their position

Engineering Contradiction:

Improveadaptability to varying individual heightsVSAvoiddevice structure

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent transitions from a standard planar camera view to a spherical coordinate system for image processing. By mapping fisheye camera images onto a spherical surface and using spherical coordinate transformations, the system can capture and process facial images from multiple angles and distances simultaneously, accommodating individuals of varying heights without requiring physical camera repositioning

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The patent employs spherical projection to map the distorted fisheye camera images onto a spherical surface. This spherical coordinate system allows the system to handle the curved field of view naturally, enabling accurate facial recognition across different viewing angles and distances, thus adapting to various individual heights and positions

Inventive Principle:

Principle #14Spheroidality (Curvature)

2Measurement precision

If fisheye camera images are converted to plane images using conventional methods, then the processing is simple, but precision is lost due to rotational symmetry errors

Engineering Contradiction:

Improvefacial recognition precisionVSAvoidimage processing complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent maintains the spherical nature of fisheye images throughout processing rather than flattening them. By using spherical coordinate systems and spherical projection methods, the system preserves the rotational symmetry properties of the original images, eliminating precision errors that would otherwise occur during planar conversion while enabling accurate facial feature extraction

Inventive Principle:

Principle #14Spheroidality (Curvature)

Solution Approach 2:

The patent replaces conventional planar image processing algorithms with spherical coordinate-based processing methods. This substitution allows the system to handle the intrinsic spherical geometry of fisheye images mathematically, maintaining precision by performing convolutions and feature extractions in the spherical domain rather than attempting to flatten the images

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Adaptability or versatility

If multiple cameras are positioned at different heights to accommodate varying individual heights, then adaptability improves, but device complexity and cost increase

Engineering Contradiction:

Improveaccommodation of varying individual heightsVSAvoidnumber of cameras

Core Design Contradiction:

Adaptability or versatilityVSQuantity of substance

Solution Approach 1:

The patent makes a single fisheye camera system universally applicable to individuals of all heights by implementing spherical coordinate-based image processing. The spherical projection method enables the same camera to accurately capture and process facial images from various distances and angles, eliminating the need for multiple cameras at different heights while maintaining adaptability across diverse user populations

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS12380730B2Authentication device, authentication method, and computer readable storage medium

Publication Date: 2025.08.05 SOFTBANK CORPORATION

AI summary

Provided is an authentication device comprising: a data acquisition unit that acquires spherical data including a video of a person taken by a fisheye camera; and an authentication performing unit that performs facial recognition performed by a spherical convolutional neural network on the spherical data. Provided is an authentication device comprising: a data acquisition unit that acquires spherical data including a video of a person taken by a fisheye camera; and an authentication performing unit that performs facial recognition performed by a convolutional neural network on a sub-image after extracting, from the spherical data, the sub-image of a portion of a face of the person, adjusting an orientation of the face of the person in the sub-image, and applying, on the sub-image, a distortion correction process that corrects distortion generated by the fisheye camera.