Camera-Based NG Detection System

Overview

An automated quality inspection system built for PT. Nihon Seiki Indonesia, using a camera and computer vision to count and verify manufactured parts on the production line — replacing manual counting with an instant, camera-based check.

Problem

• Counting parts by hand for every batch was slow and prone to human error, especially during high-volume shifts.
• Operators had no fast, objective way to confirm a tray had the exact quantity required before it moved to the next stage.

Solution

• Instant count: the operator places a tray under the camera and presses one key — the system counts every part in under a second.
• Clear pass/fail signal: green, yellow, or red status with matching sound cues, so operators don't need to interpret numbers.
• Visual verification: detected parts are outlined on screen, so operators can immediately see what the system counted.

---

Operator Flow

flowchart TD
    A[Operator opens the app] --> B[Live camera feed shown]
    B --> C[Place tray of parts under camera]
    C --> D[Press ENTER]
    D --> E{System processes & counts}
    E -->|count == target| F["GOOD (green) + success sound"]
    E -->|count < target| G["NO GOOD (red) + fail sound"]
    E -->|count > target| H["CHECK / CALIBRATE (yellow) + beep"]
    F --> I[Continue to next lot]
    G --> J[Recheck part quantity]
    H --> K[Recheck camera calibration/position]

Vision Processing Pipeline

flowchart TD
    A[Camera captures frame on ENTER] --> B[Save frame as image]
    B --> C[Pre-processing]
    C --> C1[Resize image]
    C --> C2[Gaussian blur]
    C --> C3[Bilateral filter - noise reduction]
    C --> C4[Grayscale conversion]
    C1 & C2 & C3 & C4 --> D[Segmentation & Binarization]
    D --> D1[Binary threshold]
    D --> D2[Invert / Otsu]
    D1 & D2 --> E[Contour Extraction & Filtering]
    E --> E1[Find external contours]
    E --> E2[Filter by contour area to remove noise]
    E1 & E2 --> F[Count valid contours = N]
    F --> G{Decision Logic}
    G -->|N < target| H1["Status: NO GOOD"]
    G -->|N > target| H2["Status: CHECK / CALIBRATE"]
    G -->|N == target| H3["Status: GOOD"]
    H1 & H2 & H3 --> I[Show result modal:<br/>count, status, contour overlay]

Application Showcase

Raw camera capture

The camera's live view of a tray of parts before processing.

Detected & counted result

The same tray after the system has isolated and counted each part against the target quantity.

Key Impact

• Reduced manual inspection time and human error in part counting.
• Gave operators instant, consistent pass/fail feedback on every batch.