CNN Object Detection and Tracking in Endoscopic Video
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing lesion detection methods in endoscopic videos face challenges due to artifacts such as over- and under-exposure, motion blur, and other distortions, leading to unreliable results in detecting and tracking gastric and colorectal cancer lesions.
Innovation Solution
A computer-implemented method using a convolutional neural network (CNN) for detecting and tracking target objects in captured video endoscopy, involving image data processing, detection and tracking model inputs, score enhancement operations, and matching outcomes to accurately mark target objects in real-time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a lesion detector based on machine learning models is directly applied to endoscopic videos, then the detection speed is improved, but the reliability of detection results deteriorates due to artifacts such as over- and under-exposure, motion blur, and video defocus
Solution Approach 1:
The system segments the detection process into two distinct components: a detection model that identifies potential lesions and a tracking model that verifies and refines detections across video frames. This segmentation allows each model to specialize, with the tracking model specifically designed to handle artifacts by leveraging temporal consistency, thereby improving reliability while maintaining detection speed.
Solution Approach 2:
The tracking model serves as an intermediary between the detection model and final results. It takes detection results as input and refines them by tracking objects across frames, using temporal information to filter out false positives caused by artifacts. This intermediary layer preserves detection speed while significantly improving reliability.
2Measurement precision
If multiple tracking models are used to improve tracking reliability, then the tracking accuracy is improved, but the device complexity increases
Solution Approach 1:
Multiple tracking models are merged into a unified tracking framework that processes detections collectively. Instead of running independent tracking models separately, the system combines their outputs and uses ensemble reasoning to improve tracking accuracy. This merging approach maintains improved accuracy while reducing the operational complexity of managing multiple separate models.
Solution Approach 2:
The tracking models are designed with multi-functionality to handle various artifact types (over-exposure, under-exposure, motion blur, defocus) using a unified approach. By making the tracking model universal in handling different artifact conditions, the system achieves high tracking accuracy without needing separate specialized models for each artifact type, thereby controlling device complexity.
3Reliability
If detection score enhancement operations are performed to reduce missed lesions, then the detection reliability is improved, but the processing time increases
Solution Approach 1:
The system performs preliminary filtering of detection results using the tracking model before applying enhancement operations. By pre-identifying high-confidence detections through temporal tracking, the system can focus enhancement operations only on ambiguous cases, thereby improving detection reliability while minimizing the time cost of enhancement processing.
Solution Approach 2:
The detection score enhancement is applied selectively rather than uniformly to all detections. The system applies enhancement operations only to detections that fall within a specific confidence threshold range, performing partial action on the most critical cases. This approach improves reliability for uncertain detections while avoiding unnecessary processing time on high-confidence detections.
Data Source
AI summary
A method for detecting and tracking target object in a captured video using convolutional neural network (CNN) is provided. The method includes: inputting image data into a detecting model to generate detection results, wherein the detecting model is constructed by the CNN; inputting the image data into tracking models to generate tracking results; performing detection score enhancement operation according to the detection results and the tracking results to obtain enhanced detection results; matching the enhanced detection results and the tracking results by a matching operation; processing matched results and unmatched target detection results and unmatched target tracking results; and selectively updating the tracking models using tracking reliability estimation according to the matched results.