AI-Based
Image Recognition Systems
for Industrial Machinery

【Overview】
We addressed challenges in labor reduction and quality control in logistics and manufacturing environments by leveraging high-accuracy AI-based image recognition together with multi-modal sensors (2D/3D).

From detecting the state of specific objects to identifying complex defects on manufacturing lines, we provide custom solutions tailored to a wide range of industrial use cases.

【Example 1】
Automated Detection System for Twist Lock States on Truck Container

  •  Solution Provided

    We delivered a system that automatically detects and classifies the lock/unlock state of twist locks used to secure truck containers.

    The system achieves high reliability even in harsh outdoor environments, using cameras installed at a distance of approximately one meter from the target.

  • Applied Technologies

    ・ Multi-Modal Sensing (2D + 3D)
    - Recognition accuracy using a single RGB (2D) camera was insufficient, while standalone ToF (3D) sensors showed performance degradation at distances beyond 70 cm.

    - To overcome these limitations, a combined 2D + 3D sensing approach was adopted and implemented.
    ・Multi-Camera Configuration
    - Wide-angle RGB camera: Used for precise localization of the inspection target

    - Telephoto RGB camera: Used for detailed detection of the lock state

    - 3D ToF sensor: Acquires point cloud data using a 940 nm laser, minimizing the impact of outdoor ambient light
    ・Proprietary Algorithms
    - 2D features are extracted from RGB images using CNN-based models

    - 3D features are extracted from ToF point cloud data using PointNet

    - These features are integrated using a proprietary fusion algorithm, achieving high classification accuracy and robustness

【Example 2】
Automated Detection System for Connector Connection Failures on Manufacturing Lines

  •  Solution Provided

    We automated visual inspection processes that were previously performed manually, detecting connector connection failures on printed circuit boards—such as incomplete insertion, unlocked connectors, and misaligned flexible cables.
    By combining USB cameras with AI-based analysis, the system improves inspection efficiency while reducing manual labor.

  • Applied Technologies

    ・Edge AI System
    - AI models are deployed on edge devices to enable high-speed, low-latency decision-making
    ・Proprietary AI Models

    - We developed hybrid AI models that combine region segmentation and geometric consistency, providing robustness against appearance variations, illumination changes, and reflections
    ・Operational Support Software
    - We developed and implemented software that assists operators during setup and inspection

    - The system visually indicates connector positions using white bounding frames on the camera preview, supporting accurate placement and smooth operation

Back to Development Cases

©2020 Calibur Technology Co., Ltd.

en/