Image Perspective Transformation for Foreshortening Correction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Images captured from a downward camera angle often exhibit a 'foreshortening' effect, where the subject's head and upper torso appear disproportionately large relative to the legs and lower body due to the camera's orientation, leading to distorted proportions.

Innovation Solution

The implementation of a system that applies nonlinear perspective transformation to correct for this distortion, specifically by stretching the legs and compressing features not in the same vertical plane as the rest of the body, ensuring all parts of the subject appear proportionate, using a combination of linear and nonlinear corrections.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Area of stationary object

If a downward camera angle is used to capture images, then the camera can capture the subject's full body and head, but the subject's head and upper torso appear disproportionately large relative to the legs and lower body

Engineering Contradiction:
Improvefield of viewVSAvoidbody proportions
Core Design Contradiction:
Area of stationary objectVSShape

Solution Approach 1:

The patent applies perspective transformation by changing the geometric parameters of the image coordinates. A transformation function maps each pixel position (x, y) in the original image to a new position (x', y') in the transformed image, effectively altering the spatial parameters to correct the foreshortening effect while preserving the overall composition

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent introduces an additional transformation dimension by applying a perspective transformation that simulates a different camera angle. The transformation uses a perspective divide operation (dividing by a depth factor) to create the effect of viewing the subject from a more level angle, adding a virtual dimension of depth correction

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

2Shape

If linear perspective transformation is applied to stretch the legs, then the foreshortening effect is reduced, but features not in the same vertical plane (such as feet protruding forward) appear overly stretched

Engineering Contradiction:
Improvebody proportionsVSAvoidfeature accuracy
Core Design Contradiction:
ShapeVSManufacturing precision

Solution Approach 1:

The patent applies different transformation strengths to different regions of the image. The transformation factor varies based on the vertical position and horizontal distance from the center, allowing moderate stretching for the legs while applying less stretching or compression to the feet and other protruding features, thus maintaining local accuracy while correcting overall proportions

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent uses a dynamic transformation factor that adapts to each pixel's position in the image. Rather than applying a uniform linear stretch, the transformation factor is calculated dynamically for each pixel based on its coordinates and the subject's geometry, allowing the transformation to adapt locally to preserve feature accuracy

Inventive Principle:
Principle #15Dynamics

Data Source

PatentUS10013736B1Image perspective transformation system
Publication Date: 2018.07.03 AMAZON TECH INC
  • US10013736B1 patent drawing
  • US10013736B1 patent drawing
  • US10013736B1 patent drawing

AI summary

Various examples are directed to systems, devices, and methods effective to transform an image perspective to compensate for a foreshortening effect resulting from an image taken at an angle. An image capture device may capture an image of a subject. A pitch angle of the image capture device may be determined. A transformed, stretched image may be generated by performing a linear perspective transformation of the image. At least one compression value for a portion of the transformed, stretched image may be determined based at least in part on the pitch angle. A nonlinearly transformed image may be generated based at least in part on the portion and the at least one compression value.