DBpia: https://www.dbpia.co.kr/journal/articleDetail?nodeId=NODE11722643
download: https://journal.kci.go.kr/jksci/archive/articlePdf?artiId=ART003055300
This paper introduces a method for generating model images that can identify specific cylindrical medicine containers in videos and investigates data collection techniques. Previous research had separated object detection from specific object recognition, making it challenging to apply automated image stitching. A significant issue was that the coordinate-based object detection method included extraneous information from outside the object area during the image stitching process. To overcome these challenges, this study applies the newly released YOLOv8 (You Only Look Once) segmentation technique to vertically rotating pill bottles video and employs the ORB (Oriented FAST and Rotated BRIEF) feature matching algorithm to automate model image generation. The research findings demonstrate that applying segmentation techniques improves recognition accuracy when identifying specific pill bottles. The model images created with the feature matching algorithm could accurately identify the specific pill bottles.
Keywords: Image Detection, Feature Matching, Image Stitching, YOLOv8, Image Segmentation.
Medication safety and management are critical components of modern healthcare and the pharmaceutical industry. Patients and medical professionals must systematically monitor medication intake to ensure safety and therapeutic efficacy. To address these challenges, numerous attempts have been made to integrate computer technologies into medication management systems. This study focuses on recognizing individual pill bottles to develop a system capable of effectively managing medication adherence.
Object recognition research often employs techniques such as Optical Character Recognition (OCR), image detection, and feature matching to identify specific items. In domestic studies on object detection, the performance of two-stage detectors, such as Fast-RCNN and RCNN, has been extensively compared with YOLO-based one-stage detectors [7, 8].
The advancements in YOLO versions have led to significant improvements in recognition rates, bounding box accuracy, and additional functionalities like segmentation and 3D recognition. Due to its faster detection speed in real-time video processing compared to two-stage models, YOLO has been selected as the research model in this study [13]. However, while object detection models are effective at determining whether the target is a pill bottle, they face challenges in identifying specific products.
To overcome these limitations, this study integrates a feature matching algorithm into the object detection process. After detecting the object as a pill bottle, the system analyzes its features to identify the specific product. For example, prior research [8] has explored extracting images at angles between 0° and 360° during vertical-axis rotations and comparing recognition rates based on image stitching intervals.
This study combines these approaches to create model images for specific object detection and evaluates their effectiveness. By automating the extraction of pill bottle segments from user-generated videos and stitching them into comprehensive model images, this research achieves enhanced object recognition.
The proposed system leverages YOLOv8 for segmenting pill bottles and employs the ORB algorithm for feature matching to identify specific objects. To improve feature matching accuracy, the system replaces bounding box extraction with segmentation during object detection. Furthermore, the ORB algorithm automates image clustering for stitching. This approach not only accelerates pill bottle detection via cameras but also increases recognition rates, efficiently collects image data, and automates medication management tasks.
The structure of this paper is as follows: Section 2 introduces the core technologies and models used in the system and reviews related studies on specific object detection. Section 3 details the proposed system and experimental methods. Section 4 presents the results of each process. Section 5 concludes with a discussion of limitations, suggested improvements, and future research directions.