Object Classification via Masked 2D Projection Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current baggage screening techniques struggle to accurately identify sub-objects within compound objects using CT scanners, leading to reduced throughput and false positives or false negatives, as existing methods like compound object splitting can indiscriminately split objects or fail to separate compound objects effectively.
Innovation Solution
A method and system that classify potential compound objects without image data segmentation, using fidelity scores and masking techniques to isolate and compare characteristics of sub-objects by generating a masked projection that excludes outliers, allowing for accurate classification of sub-objects as potential threats or non-threats.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If compound object splitting is used to separate sub-objects, then threat detection accuracy is improved, but false positives increase and unnecessary inspections occur
Solution Approach 1:
The patent applies segmentation by dividing the compound object into multiple sub-objects based on detected boundaries and characteristics. The system identifies individual items within the compound object (e.g., separating a knife from a book) and evaluates each sub-object independently against threat criteria, thereby improving detection accuracy while reducing false positives through targeted rather than universal splitting.
Solution Approach 2:
The patent implements local quality by applying different processing strategies to different regions of the compound object. Instead of uniform splitting, the system identifies specific sub-objects with threat-like characteristics and applies detailed analysis only to those regions, while treating other regions differently. This localized approach maintains high accuracy for potential threats while reducing overall false positive rates.
2Shape
If erosion is applied to split compound objects, then object separation is achieved, but object mass is reduced and non-compound objects are incorrectly split
Solution Approach 1:
The patent applies preliminary action by first analyzing the compound object to identify actual boundaries and sub-object characteristics before performing any separation. The system detects edges, contours, and material property variations to determine where genuine separations exist, then applies splitting only at these identified boundaries. This preliminary identification prevents incorrect splitting of non-compound objects while achieving proper separation where needed.
Solution Approach 2:
The patent inverts the conventional approach by not starting with erosion and seeing what separates, but rather starting with identification of actual object boundaries and then applying separation only where justified. Instead of applying a universal erosion algorithm that may incorrectly split objects, the system first identifies genuine sub-object boundaries through multiple analysis methods, then performs targeted separation only at these validated locations.
3Adaptability or versatility
If universal erosion and splitting is applied to all objects, then processing consistency is maintained, but throughput decreases due to unnecessary processing
Solution Approach 1:
The patent applies partial action by performing detailed compound object splitting and analysis only on objects that exhibit characteristics suggesting they are compound objects. The system uses initial screening to identify potential compound objects, then applies comprehensive splitting and sub-object analysis only to these candidates. Objects that do not show compound characteristics undergo simpler, faster processing, thereby maintaining consistency for compound objects while improving overall throughput by avoiding unnecessary processing of simple objects.
Data Source
AI summary
One or more systems and/or techniques are provided to identify objects comprised in a compound object without segmenting three-dimensional image data of the potential compound object. Two-dimensional projections of a potential compound object (e.g., Eigen projections) are examined to identify the presence of known objects. The projections are compared to signatures, such as morphological characteristics, of one or more known objects. If it is determined based upon the comparison that there is a high likelihood that the compound object comprises a known object, a portion of the projection is masked, and it is compared again to the signature to determine if this likelihood has increased. If it has, a sub-object of the compound object may be classified based upon characteristics of the known object (e.g., the compound object may be classified as a potential threat item if the known object is a threat item).