X-ray detector pose estimation using occluded marker inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual alignment of X-ray detectors and sources in robotic X-ray systems is inconsistent, time-consuming, and prone to inaccuracies due to occluded or out-of-view markers, limiting the quality of X-ray images.
Innovation Solution
A machine-learned model is used to estimate the pose of X-ray detectors by identifying visible and occluded markers from images, even when they are not directly visible, allowing for automatic alignment of the X-ray source with the detector, using a combination of deep neural networks to localize and refine marker positions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If manual alignment is used to position the X-ray detector and source, then the system provides flexibility in imaging, but the alignment quality becomes inconsistent and time-consuming
Solution Approach 1:
The patent replaces manual mechanical alignment operations with an automated computer vision system. A camera captures images of the detector, and software automatically identifies markers and calculates detector pose and alignment parameters, eliminating the need for manual measurement and positioning while maintaining flexibility.
Solution Approach 2:
The system enables self-alignment by using the detector's own markers as reference points. The automated detection system processes images of the detector markers to determine detector position and orientation, allowing the system to self-correct and self-align without external intervention.
2Productivity
If automatic alignment using marker detection is implemented, then alignment speed improves, but accuracy deteriorates when markers are occluded or out-of-view
Solution Approach 1:
The patent performs preliminary actions by capturing multiple images of the detector from different angles and positions before the actual alignment process. This creates a library of marker positions that can be referenced even when some markers are occluded during the alignment operation, ensuring continuous accuracy.
Solution Approach 2:
The patent introduces an intermediary computational model that predicts the positions of occluded or out-of-view markers based on visible markers and the detector's known geometry. This virtual marker generation acts as a mediator between visible markers and the complete set of markers needed for accurate pose estimation.
3Ease of manufacture
If hand-crafted filters and Hough transform are used for marker detection, then the detection process is straightforward, but the system struggles with large distance variations and occlusions
Solution Approach 1:
The patent changes the detection parameters dynamically based on imaging conditions. The system adjusts marker detection thresholds, search regions, and filtering criteria according to the detected distance variations and occlusion levels, allowing the same system to handle diverse imaging scenarios effectively.
Solution Approach 2:
The patent transforms the static marker detection approach into a dynamic system that adapts to varying conditions. The detection algorithm continuously adjusts its parameters based on real-time image quality, marker visibility, and distance measurements, making the system flexible rather than rigid.
Data Source
AI summary
For x-ray detector pose estimation, a machine-learned model is used to estimate locations of markers, including occluded or other non-visible markers, from an image. The locations of the markers, including the non-visible markers are used to determine the pose of the X-ray detector for aligning an X-ray tube with the X-ray detector.