Smart Image Tagging Model for Autonomous Vehicle Damage Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for assessing vehicle damage, such as those used for insurance claims, are labor-intensive and rely on manual image capture and review by experts, leading to variability in image quality and completeness.
Innovation Solution
A mobile device equipped with a smart image tagging model that automatically captures, tags, and selects images of a vehicle, predicting tags associated with different vehicle portions and transmitting representative images to a back-end server without user intervention, using a light-weight convolutional neural network model trained on depth-wise separable and residual convolutions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual image capture and review by experts is used, then image quality and completeness can be assessed, but the process is labor-intensive and cost-intensive
Solution Approach 1:
The system automatically captures images, tags them with vehicle portion identifiers, and selects representative images without requiring expert intervention. The mobile device performs self-service by autonomously completing the entire image collection and selection process, eliminating the need for manual expert review while maintaining high image quality standards.
Solution Approach 2:
The patent replaces the mechanical system of manual expert image review with an automated computer vision system. The smart image tagging model and frame processor automatically analyze and select images, substituting human expertise with algorithmic processing to improve efficiency while maintaining assessment quality.
2Loss of information
If prompts are provided to guide users to capture specific images, then image completeness improves, but user active participation is required
Solution Approach 1:
The system eliminates the need for user interaction by automatically capturing images and tagging them with vehicle portion identifiers. The mobile device performs self-service by autonomously determining which images are needed and capturing them without requiring users to follow prompts or actively participate in the image collection process.
Solution Approach 2:
The patent extracts the cognitive burden of understanding prompts and selecting images from the user and transfers it to the automated system. The smart image tagging model handles all decision-making about which images to capture and how to tag them, completely removing the need for user active participation while ensuring image completeness.
3Measurement precision
If back-end server is used to determine photo acceptability, then image quality control improves, but processing time and server load increase
Solution Approach 1:
The patent segments the image processing workflow by performing all quality control and selection operations locally on the mobile device using the frame processor and smart image tagging model. This divides the overall process into local automated processing (image capture, tagging, and selection) and minimal server communication (transmitting only selected representative images), thereby reducing server load and processing time while maintaining quality control.
Solution Approach 2:
The system performs preliminary image processing and selection actions locally on the mobile device before transmitting images to the server. The frame processor pre-selects representative images based on tagging and confidence scores, so only the most relevant images are sent to the server, significantly reducing transmission time and server processing requirements while maintaining quality standards.
4Measurement precision
If multiple images are captured for each vehicle portion, then selection accuracy improves, but data transmission volume increases
Solution Approach 1:
The patent extracts only the essential representative images from the captured set based on tagging confidence scores and vehicle portion importance. The frame processor identifies and extracts only the most relevant images for each vehicle portion, eliminating redundant data transmission while preserving selection accuracy by transmitting only the highest-quality representative images.
Solution Approach 2:
The system applies local quality assessment by evaluating each image's tagging confidence score and vehicle portion significance individually. Rather than transmitting all captured images uniformly, the system selectively transmits only those images with high confidence scores for critical vehicle portions, optimizing data transmission volume while maintaining selection accuracy through localized quality filtering.
Data Source
AI summary
Techniques for automatic image tagging and selection at a mobile device include generating a smart image tagging model by first training an initial model based on different angles of image capture of subject vehicles, and then re-training the trained model using weights discovered from the first training and images that have been labeled with additional tags indicative of different vehicle portions and/or vehicle parameters. Nodes that are training-specific are removed from the re-trained model, and the lightweight model is serialized to generate the smart image tagging model. The generated model may autonomously execute at an imaging device to predict respective tags associated with a stream of frames; select, capture and store respective suitable frames as representative images corresponding to the predicted tags; and provide the set of representative images and associated tags for use in determining vehicle damage, insurance claims, and the like.