Single Camera Dimension Measurement Using Deep Learning Normalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current dimension measurement technologies, such as 2D line laser, 3D line laser, and structured light sensors, face limitations in accuracy, cost, and environmental adaptability, particularly when measuring complex objects or large areas, and suffer from decreased accuracy due to aging sensors.
Innovation Solution
A method and apparatus using a single camera with a deep learning model to estimate dimensions from a 2D image, involving image normalization and denormalization using pre-trained models, and depth information to generate accurate dimension information, allowing for training across various environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a 2D line laser is used for dimension measurement, then the cost is reduced, but the measurement capability is limited to small areas and cannot measure large areas effectively
Solution Approach 1:
The patent transitions from 2D line laser measurement to 3D point cloud measurement by introducing depth information through a camera. This dimensional change enables measurement of large areas while maintaining cost-effectiveness, as the camera-based approach captures three-dimensional spatial data without requiring multiple 2D laser scanners or mechanical movement systems
2Measurement precision
If a 3D line laser is used for dimension measurement, then the measurement accuracy for large areas is improved, but the cost increases due to the need to move the stand or line laser
Solution Approach 1:
The patent employs a camera that serves multiple functions: capturing images for dimension measurement, providing depth information through stereo vision or focus analysis, and enabling both close-up and distant measurements. This multi-functional approach replaces the need for expensive, movable 3D line laser systems while maintaining measurement accuracy across various distances and areas
3Measurement precision
If a structured light sensor is used for dimension measurement, then high measurement accuracy is achieved, but the accuracy decreases when the target object has a complex structure such as holes
Solution Approach 1:
The patent replaces the structured light projection mechanism with a passive camera-based depth sensing approach. By using stereo vision or focus-based depth estimation, the system captures depth information without requiring active light projection, thereby avoiding the problem of light being blocked or scattered by complex object structures like holes, while maintaining measurement accuracy
4Ease of operation
If stereo or multi-stereo cameras are used for dimension measurement, then dimension measurement is enabled, but the accuracy decreases as the cameras age
Solution Approach 1:
The patent incorporates a feedback mechanism where the system continuously calibrates and adjusts the depth estimation parameters based on captured images and known reference dimensions. This feedback loop compensates for aging effects in the camera sensors, maintaining measurement accuracy over time without requiring physical replacement or recalibration of the camera hardware
Data Source
AI summary
Disclosed are an apparatus and method for measuring a dimension. The method of measuring a dimension according to an aspect of the present invention includes photographing a target object and generating an image, normalizing the image of the target object, generating normalized estimated dimension information about the target object from the normalized image of the target object using a pre-trained dimension prediction model, and converting the normalized estimated dimension information of the target object into estimated dimension information of the target object.