Real-Time Marker Tracking via Shape Segmentation and SVM
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems face challenges in accurately and efficiently tracking a marker in real-time, particularly in healthcare applications where precise head motion analysis is required, due to random movement patterns and high processing demands.
Innovation Solution
A monocular tracking scheme that uses a green circular active color marker mounted on the subject's forehead, employing shape-based segmentation, color filtering, and a support vector machine (SVM) model for real-time detection and tracking, optimizing processing time while maintaining high precision and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional video-based object tracking methods are used, then the system can track objects in video, but the processing time increases and real-time performance deteriorates when dealing with huge volume of video data
Solution Approach 1:
The patent segments the video processing task into distinct phases: detection phase (identifying marker in current frame) and tracking phase (predicting marker location in subsequent frames). This segmentation allows the system to process only relevant regions in each phase, reducing overall processing time while maintaining real-time performance.
Solution Approach 2:
The system performs preliminary actions by establishing a representation of the target marker from initial detection, determining dominant motion trajectory characteristics in advance, and using these to predict future positions. This preliminary action reduces processing time for subsequent tracking operations.
2Measurement precision
If the marker detection sensitivity is increased to capture subtle behavior details, then the measurement precision improves, but the processing complexity increases
Solution Approach 1:
The patent applies local quality by focusing computational resources on the region of interest (marker location) rather than processing the entire video frame uniformly. The system establishes a representation of the target and tracks only the relevant marker, reducing processing complexity while maintaining high detection accuracy for subtle behavior details.
3Measurement precision
If the tracking system processes the entire video frame-by-frame, then the detection accuracy is maintained, but the productivity decreases due to huge volume of data
Solution Approach 1:
The patent extracts only the essential information needed for tracking by establishing a representation of the target marker and using dominant motion trajectory characteristics. Instead of processing the entire video frame-by-frame, the system extracts and tracks only the marker's position and motion patterns, significantly improving processing efficiency while maintaining tracking accuracy.
4Device complexity
If the system uses a monocular camera for wearable marker tracking, then the device complexity is reduced, but the measurement precision deteriorates due to random movement patterns
Solution Approach 1:
The patent applies dynamics by adapting the tracking representation based on dominant motion trajectory characteristics. The system dynamically adjusts to random movement patterns by using motion trajectory information to predict and compensate for marker position changes, maintaining measurement precision with a simple monocular camera setup.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
Systems and methods for detecting and tracking a marker in real time are disclosed. Shape based segmentation of at least one object detected in a first frame from a sequence of frames is performed to define a region of interest (ROI) surrounding an object of interest corresponding to the marker. A marker detection model is dynamically trained based on sampling points from a plurality of pixels in and around the ROI. The marker is then tracked in real-time based on projected ROI in subsequent frames and the trained marker detection model. To optimize computation time required in classifying the pixels as marker pixels or non-marker pixels, the ROI is reduced to half its size, classification is performed on the reduced ROI and to improve accuracy, blob detection and classifying pixels along the boundary of the reduced ROI is performed by processing the ROI in original resolution.