Why you should opt
Konan Watcher
We recommend Konan Watcher to all business areas requiring all-in-one workflow
for video collection, analysis, processing, tagging, and learning.
Konan Watcher identifies various multi-modals, such as faces, objects, situations,
and scenes within video content, and works as a solution that can be applied to diverse industries,
including public areas, national defense, broadcasting, and cultural art.
BENEFITS
Benefit 01.
Data Securing Capabilities That Make You Stand Out
With its super-resolution (SR) technology based on GAN, Konan Watcher generates high-quality images, tracks videos, and secures high-quality data utilizing 3D mapping technology.
Benefit 02.
Easy Data Collection, Refinement, Learning, and Evaluation
Konan Watcher provides a systematic management feature by utilizing its user-friendly UI and UX screens and automating all data collection, refinement, learning, and evaluation procedures.
Benefit 03.
Immediate Anomaly Detection and Countermeasures
Based on the convergence between the digital video management platform and deep learning platform, Konan Watcher enables the placement and real-time processing of large-volume videos, making it suitable for areas demanding immediate situation detection and response, such as surveillance camera supervision, smart factories, and contactless commerce.
FEATURE
By combining various AI technologies, such as large-volume video processing technology, including high-speed conversion and transmission, image division in the unit of frames, and metadata extraction, Konan Watcher provides Korea’s highest level of object recognition and detection performance.
Video Compression and Conversion/Frame Extraction
- Video conversion feature for various formats such as SD/HD/UHD
- Scene conversion precision provided in the unit of frames
- Extracts points for cut, wide, and fade conversions
- Extracts key frames
- Scene conversion recognition and sensitivity adjustment features
Clustering and De-identification Processing
- Conducts clustering that extracts characteristics from multiple unrecognized facial images and groups similar ones together as well as efficient labeling work through the clustering process
- Utilizes face detection and recognition technologies to non-differentiate all characters except specific individuals inside the video
Super Resolution (SR)
- Scale down (high resolution → low resolution):
low-capacity conversion to reduce rendering time - Scale up (low resolution → high resolution):
converting low-resolution sources to high-resolution - Low-resolution video → Video separation → Conversion video into images (cataloging) → SR application to converted images
Data Augmentation
- A wide range of data augmentation methods (crop, flip, resize, rotate, brightness, diffuse, blur, sharpen, etc.) are used to generate various pieces of data to supplement insufficient learning data
- Various data cases are created through a wide range of deep learning GAN (CycleGAN, WGAN-GP, StyleGAN, etc.) algorithms supplement insufficient learning data
-
Facial recognition models based on deep learning, such as FaceNet, are applied to provide face detection, characteristics extraction, classification, and clustering features.
-
Deep learning-based object recognition models, such as Yolo types and object division, are applied to provide object detection, characteristics extraction, classification, and clustering features.
-
We provide an all-in-one workflow feature for video collection, analysis, processing, tagging, and learning.
-
Various Generative Adversarial Network (GAN) technologies are used to secure learning data.
-
Through domain-specialized face and object recognition features, the solution provides optimal recognition services.
-
We provide the Restful API that can be integrated with various systems, including Video Management System (VMS) and drones.
TECHNOLOGY
USE CASES
Through Konan Watcher’s deep learning-based facial recognition and clustering technologies, primary and proximal individuals appearing in the video are automatically tagged, and unlearned individuals are automatically clustered, After which the time and costs related to reorganizing person-related images and videos collected by the thousands over a day are significantly reduced.
CASES01. Improvement in the Efficiency of Presidential Archives’ President Surroundings Data Organization Work
Through Konan Watcher’s deep learning-based facial recognition and clustering technologies, the Presidential Archives automatically tagged crucial and secondary individuals appearing in videos and automatically clustered unlearned individuals. Through this process, the agency has been able to significantly reduce the time and costs related to reorganizing individual-related images and videos collected by the thousands over the course of a day through the president’s surroundings.
CASES02. ROK Army Training and Doctrine Command Contributes to Fielding Military Data by Applying AI Convergence Technology
The Korean Army established Military ImageNet (2020–2021) as a video database and is constructing the Intelligence Center (2023–2025) as a database platform to develop AI core capabilities. As a part of its efforts, the Army used Konan Watcher to automate a series of procedures, including learning data establishment, learning model establishment, and learning model distribution. In addition, as there are limitations to AI learning due to the constraints on securing learning data in areas of national defense, Konan Watcher provided effective augmentation techniques for small amounts of data that had been secured.
