Teacher Data Generation via Coordinate Transformation

Solution Overview

Problem

Generating teacher data for image recognition from a large number of diverse viewpoints is labor-intensive and costly, requiring a significant number of images.

Innovation Solution

A method involving capturing images from designated positions, extracting basic image regions, and applying coordinate transformations to generate teacher data as if the images were taken from different positions, reducing the number of images needed.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If teacher data is generated from a large number of captured images taken from different viewpoints, then the quality and comprehensiveness of image recognition training data is improved, but the man-hour and cost required for data generation considerably increase

Engineering Contradiction:

Improvequality of teacher dataVSAvoidman-hour for data generation

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent creates virtual captured images by copying and transforming a basic image region from one actual captured image through coordinate transformations. Instead of manually capturing images from all desired viewpoints, the system generates synthetic images that simulate different camera positions and orientations, thereby reducing manual data collection effort while maintaining data quality for training purposes

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent performs preliminary extraction of the basic image region from actual captured images before generating the full set of teacher data. By pre-processing and identifying the essential object region in advance, the system prepares data in a format that enables efficient subsequent coordinate transformations and virtual image generation, reducing overall processing time

Inventive Principle:

Principle #10Preliminary action

2Reliability

If teacher data is generated from a large number of captured images taken from different viewpoints, then the comprehensiveness of image recognition training data is improved, but the cost of data generation considerably increase

Engineering Contradiction:

Improvecomprehensiveness of teacher dataVSAvoidcost of data generation

Core Design Contradiction:

ReliabilityVSLoss of energy

Solution Approach 1:

The patent creates virtual captured images by copying and transforming a basic image region from one actual captured image through coordinate transformations. Instead of manually capturing images from all desired viewpoints, the system generates synthetic images that simulate different camera positions and orientations, thereby reducing manual data collection effort while maintaining data quality for training purposes

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The system uses the basic image region and coordinate transformation information to automatically generate virtual captured images without requiring additional manual image capture or processing. The transformation unit self-generates the necessary teacher data by mathematically transforming the basic region according to predetermined camera positions, eliminating the need for external resources

Inventive Principle:

Principle #25Self-service

3Loss of time

If the number of images used as basis for generating teacher data is reduced, then the man-hour and cost are reduced, but the sufficiency of teacher data for image recognition may be compromised

Engineering Contradiction:

Improveman-hour for data generationVSAvoidsufficiency of teacher data

Core Design Contradiction:

Loss of timeVSReliability

Solution Approach 1:

The patent transitions from capturing images in physical three-dimensional space to generating images through mathematical coordinate transformations in a transformed coordinate system. By using predetermined camera position information and performing coordinate transformations on the basic image region, the system generates virtual images from multiple viewpoints without physically moving the camera, thus maintaining data sufficiency while reducing the number of actual captures needed

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The patent creates virtual captured images by copying and transforming a basic image region from one actual captured image through coordinate transformations. Instead of manually capturing images from all desired viewpoints, the system generates synthetic images that simulate different camera positions and orientations, thereby reducing manual data collection effort while maintaining data quality for training purposes

Inventive Principle:

Principle #26Copying

Data Source

PatentUS10311317B2Teacher data generation method

Publication Date: 2019.06.04 HONDA MOTOR CO LTD

AI summary

Provided is a method for generating teacher data for image recognition while reducing the number of images used as the basis. A captured image is obtained by imaging an object by an imaging device C arranged at a first designated position Pi. A basic image region Si is extracted from the captured image. The teacher data is generated as a result of coordinate transformation of the basic image region Si from one image coordinate system to a coordinate system of a captured image by the imaging device C on the assumption that the imaging device C is arranged at a second designated position Pj which is different from the first designated position Pi.