Pavement Condition Rating via Rectangle Count Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pavement condition rating methods using road surface images face challenges in accuracy due to perspective distortions, limited usable image areas, and potential for erroneous damage detection, especially in non-flat road sections, and require time-consuming calibration and continuous angle adjustments.
Innovation Solution
A pavement condition rating method that utilizes a combination of image analysis and machine learning to detect damage using an object detection model, where detected damage is surrounded by rectangles, and combines visual inspection results to build a rating estimation model, allowing for comprehensive image area usage and reduced susceptibility to erroneous detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pixel area is used to detect damage in road surface images based on perspective, then damage detection can be performed, but the accuracy is reduced because pixel area varies with distance from the shooting point
Solution Approach 1:
The patent changes the parameter used for damage detection from pixel area to the number of rectangles surrounding detected damage objects. This parameter transformation eliminates the influence of perspective distortion and distance variations, as the rectangle count remains consistent regardless of the actual size or distance of the damage in the image.
2Measurement precision
If only damage near the shooting point is detected, then perspective distortion is minimized, but the usable image area becomes limited and curved road sections cannot be analyzed
Solution Approach 1:
By changing the detection parameter from pixel area to rectangle count, the patent enables the entire image area to be utilized for damage detection, including curved road sections and areas far from the shooting point, without being constrained by perspective distortion issues.
3Reliability
If the angle of view is fixed to reduce erroneous detection, then calibration time and effort are required and continuous angle adjustments are needed for non-uniform road slopes
Solution Approach 1:
The patent changes the detection parameter to rectangle count, which is invariant to angle of view changes. This eliminates the need for calibration and continuous angle adjustments, as the rectangle count method automatically adapts to varying road slopes and camera angles without requiring manual intervention.
4Adaptability or versatility
If the angle of view changes continuously during shooting, then adaptation to road slope variations is possible, but the relationship between photographed area and pixel area becomes unpredictable causing erroneous detection
Solution Approach 1:
By using rectangle count instead of pixel area, the patent maintains adaptability to road slope variations and angle of view changes while eliminating the measurement precision problems. The rectangle count parameter is inherently invariant to these changes, allowing the system to handle diverse road conditions without compromising detection accuracy.
Data Source
AI summary
By using a plurality of images obtained by combining image of damage included in a road surface still image based on perspective and definitions of the damage as damage learning data, an object detection model is built. A plurality of data wherein the number of rectangles in an analysis result image displaying the rectangles surrounding the damage output as an analysis result of the object detection model and a pavement condition rating result obtained by visual inspection of the analysis result image of the object detection model are combined is used as condition rate learning data to build a rating estimation model, and pavement condition is rated using the rating estimation model.