Development/Research Completed
Vision Intelligence

Development of image acquisition methods and devices

  • Development of Field Image Acquisition System
  • Few-Shot Learning Based on EDANet Learning Data Augmentation
  • Development of Learning Data Construction Technology for Construction Sites

Development of Deep Learning-Based Risk Identification Algorithm

  • Target Image Learning and Analysis
  • Development and optimization of risk discrimination algorithms
  • Semi-Supervised Learning to Minimize Labeling Operations

Development of Visual Intelligence-Based Safety Management System

  • Field monitoring and risk control systems
  • Development of field risk prediction techniques
  • An Analysis of Accident Cases by Analyzing Risk Factors in the Field
  • Development of image recognition target derivation process

Development/Research Plan
Vision Intelligence

Stereo image and deep learning algorithm based personnel positioning

  • Stereo camera images are used to track the three-dimensional position of personnel or equipment
  • Replace object recognition and tracking algorithms with deep learning-based algorithms
  • Simplify the matching process between objects recognized in each of the two images
  • Comparative review of using 2 and 3 cameras
스테레오 카메라 기반 위치 추적 기술 개념도

Conceptual Diagram of Stereo Camera-Based Positioning Technology

두 대의 카메라에서 추적(2D Tracking)된 결과

Results tracked (2D tracking) from two cameras

두 대의 카메라에서 매칭(Entity Matching)된 결과

Results Matched on two cameras

Development of Data Learning Algorithms and Service Applications for Construction Sites

  • Development of data learning algorithm selection model
    Development of optimal learning algorithms for each safety management scenario (R-CNN, Fast R-CNN, SSD, YOLO, etc.) and optimization of learning algorithms and setting of deep neural network hyperparameters by safety management scenario
  • Development of Virtual Space-Data Augmentation Integration Module to Prevent Data Learning Overfitting
    Development of data augmentation module to prevent overfitting by integrating the virtual construction environment generation model and the functions of mirroring, random cropping, rotation, sharing, local wrapping, and color shifting
  • Developing safety management service applications
    Development of a support model for creating a modular construction safety work scenario and construction safety services linked to a mobile app or project management information system (PMIS)

Development of Progressive Work and Risk Inference Technology through Object Identification

If objects such as green safety helmets, workers, AL-BEAM, AL-FORM, and anti-condensation insulation are determined among the object information identified in the image, the installation of the alform slab can be inferred. If the image determines the red safety helmet, operator, sling belt, rebar, and slab, it can be inferred that rebar lifting work is in progress. Advanced object discrimination techniques enable detailed tasks to be inferred. Object information and spatial information included in the work content are subdivided to infer various work situations and analyze the types of accidents that may occur.

객체 식별 정보 기반 진행 중 작업과 발생가능 위험 분석

Analysis of ongoing operations and possible risks based on object identification

위험상황에 대한 Visual Reasoning 기술 적용

Application of Visual Reasoning Technology to Dangerous Situations