Vehicle Image Comparison Segmentation for Inspection Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle image analysis methods for inspection are cumbersome, inefficient, and error-prone, particularly in identifying physical changes while accounting for variations in imaging conditions.
Innovation Solution
A computerized method and system for vehicle image comparison that segments input images into mechanical components, retrieves reference images, and generates difference maps using deep learning models to identify probability of physical changes while excluding false alarms due to varying imaging conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual inspection with image capture is used, then visual evidence of inspection is provided, but the process is cumbersome and time-consuming
Solution Approach 1:
The patent replaces manual mechanical inspection with an automated computerized vision system that captures images and uses image processing algorithms to automatically detect and analyze vehicle defects, eliminating the need for manual visual inspection while maintaining inspection accuracy
Solution Approach 2:
The system enables self-inspection by automatically processing captured images through multiple analysis stages (pre-processing, feature extraction, defect detection, classification) without requiring human intervention, allowing the inspection system to perform its own analysis and generate results autonomously
2Measurement precision
If image registration is used for analyzing vehicle images, then comparison with reference images is enabled, but the method is inefficient, error-prone and computationally costly
Solution Approach 1:
The patent divides the vehicle image into multiple segments or regions of interest, comparing each segment independently against corresponding reference segments. This segmentation approach reduces the computational complexity of full-image registration while improving comparison accuracy by focusing on specific defect-prone areas
Solution Approach 2:
The system performs pre-processing operations on captured images before main comparison, including noise reduction, normalization, and feature enhancement. This preliminary action prepares images for more efficient and accurate comparison, reducing errors and computational burden in subsequent processing stages
3Measurement precision
If deep learning models are used for segmenting and comparing images, then accuracy in identifying physical changes is improved, but computational complexity increases
Solution Approach 1:
The patent applies segmentation to divide complex image analysis into manageable parts, using deep learning models only on specific segments rather than entire images. This reduces the computational load and model complexity while maintaining high accuracy in detecting physical changes in critical areas
Solution Approach 2:
The system applies deep learning models selectively only where needed for defect detection, rather than processing entire images with complex models. This partial application of advanced techniques maintains high detection accuracy for physical changes while reducing overall system complexity and computational requirements
Data Source
Figure 1
Figure 2
Figure 3
AI summary
There are provided a system and method of vehicle image comparison, the method including: obtaining an input image capturing at least part of a vehicle; segmenting the input image into one or more input segments corresponding to one or more mechanical components; retrieving a set of reference images, thereby obtaining a respective set of corresponding reference segments for each input segment; and generating at least one difference map corresponding to at least one input segment, comprising, for each input segment: comparing the input segment with each corresponding reference segment thereof using a comparison model, giving rise to a set of difference map candidates each indicating probability of presence of DOI between the given input segment and the corresponding reference segment; and providing a difference map corresponding to the given input segment according to probability of each difference map candidate in the set of difference map candidates.