Technology

VISION

VISION

By extracting and analyzing meaningful information from multimedia data captured by diverse vision-based sensors—including smartphones, cameras, CCTV, RGB-D devices, and LiDAR—we leverage a comprehensive suite of Vision AI techniques, spanning classical image processing to advanced classification, detection, and segmentation, to achieve specific, impactful outcomes.

Image Classification

  • Real-time image classification using AI models such as ResNet and MobileNet
  • Developed models that go beyond broad categories (e.g., cars, trucks, people) to identify vehicle details like year and model
  • •Improved accuracy and trust through Explainable AI (XAI)–driven model analysis
객체 감지

Object Detection

  • Real-time inference of object locations (bounding boxes) and classes
  • AI model well-suited for a wide range of tasks, offering fast processing and high accuracy
  • Object tracking to estimate trajectory, speed, and unique IDs
  • Performance enhancement through various data augmentation techniques

Image Segmentation

  • Technology that learns and infers image information at the pixel level
  • The most computationally intensive Vision AI analysis technique, demanding high-performance computing
  • Capable of inferring precise building outlines and window positions at the pixel level from aerial or street-view imagery
영상 처리

Image Processing

  • Enhance low-light images using histogram equalization
  • Extract specific-colored objects to obtain targeted information
  • Perform data augmentation and boost AI model performance by adjusting brightness, saturation, and contrast

On-Device AI

  • Apply various model optimization techniques such as TensorRT and ONNX
  • Enable fast inference in a Linux OS environment
  • Maximize space efficiency by using compact devices compared to desktop PCs
  • Run directly on embedded boards without needing a separate AI server