Object Detection via Parallax Correction and Reliability Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection systems face challenges in accurately detecting objects across images captured by multiple imaging devices due to parallax issues, requiring costly special devices and struggling to completely correct positional differences between images.
Innovation Solution
An object detection system that transforms images to align them, calculates reliability based on positional differences, generates integrated images, extracts feature amounts, and calculates an overall score for accurate object detection without the need for special devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple imaging devices are used to capture images in different wavelength regions, then object detection accuracy is improved, but parallax between images causes positional differences that deteriorate detection precision
Solution Approach 1:
The patent introduces a third imaging device that captures images in a third wavelength region to serve as an intermediary reference. This third image is used to correct positional differences between the first and second images by calculating displacement amounts based on correlation between the third image and the other two images, thereby eliminating parallax effects without requiring complex calibration between the primary imaging devices
Solution Approach 2:
The patent creates a corrected version of the second image by applying displacement amounts derived from the third image. This corrected second image effectively copies the positional information from the first image while retaining the wavelength region characteristics of the second image, allowing for accurate object detection across different spectral regions
2Manufacturing precision
If special devices are used to eliminate parallax influence, then positional alignment is improved, but system cost and complexity increase significantly
Solution Approach 1:
The patent enables the imaging system to self-correct positional differences by using the third imaging device to automatically calculate and apply displacement amounts. The system performs its own calibration and alignment without requiring external special devices or complex manual intervention, thereby reducing system complexity while maintaining high positional alignment precision
Solution Approach 2:
The patent replaces mechanical alignment systems with computational image processing methods. Instead of using physical calibration devices or mechanical adjustment mechanisms, the system uses correlation-based calculations and digital image transformation to achieve precise positional alignment, thereby reducing hardware complexity
3Reliability
If images are transformed to align positions, then positional difference is reduced, but complete correction is difficult to achieve
Solution Approach 1:
The patent applies displacement amounts that are calculated to partially correct positional differences by referencing the third image. Rather than attempting perfect alignment, the method applies sufficient correction to eliminate the harmful effects of parallax for object detection purposes, recognizing that complete pixel-perfect alignment is unnecessary and potentially unachievable
Data Source
AI summary
The purpose of the present invention is to detect an object in images accurately by means of image recognition without using a special device for removing the influence of the parallax between a plurality of images. An image transformation unit (401) transforms a plurality of images acquired by an image acquisition unit (407). A reliability level calculation unit (402) calculates a level of reliability representing how small the misalignment between images is. A score calculation unit (405) calculates a total score taking into account both an object detection score based on a feature quantity calculated by a feature extraction unit (404), and the level of reliability calculated by the reliability level calculation unit (402). An object detection unit (406) detects an object in the images on the basis of the total score.