Machine Vision For Quality Inspection

Problem Domain

In the context of the automated toy inspection system, the primary objective is to enhance the efficiency and accuracy of the assembly line by employing computer vision techniques to inspect and classify toys. The system focuses on the inspection of toy heads, particularly targeting the eyes, with the goal of detecting and categorizing defects in real-time.

Components and Functionalities:

  • Assembly Line Integration: Toys are placed on the assembly line for inspection. The depth camera, positioned above the line, captures images of the toys as they pass through.
  • Segmentation of Toys: Utilizing computer vision algorithms, the system performs segmentation to identify and isolate individual toys in the captured images.
  • Classification of Toy Types: Following segmentation, the system classifies the type of each toy head. Initially, the focus is on toy heads, and future expansions may include other body parts.
  • Defect Detection in Eyes: Specialized attention is given to the eyes of each toy head for defect detection.
    • Paint Area Analysis: The system analyzes the paint distribution in the eyes, distinguishing between the pupil and iris regions.
    • Alignment Check: Detects misalignments in the positioning of the eyes.
    • Gaze Direction Analysis: Determines the direction in which each eye is looking.
  • Real-time Defect Visualization: Detected defects are highlighted and visualized in real-time, providing immediate feedback to the quality control team. Visualization includes graphical overlays indicating areas of concern, misalignments, or deviations from the expected characteristics.
  • Adaptability to Changing Requirements: The system is designed with flexibility to accommodate changes in the toy production line or additional inspection criteria. Updates to the defect detection algorithms and classification models can be seamlessly integrated.
  • Integration with Production Management Systems: Inspection results, including defect details and classified toy types, can be integrated into broader production management systems. Reports and analytics may be generated for performance monitoring and process optimization.
  • Continuous Learning: The system may incorporate machine learning techniques for continuous improvement. Over time, the model learns from new data, improving its ability to accurately classify toys and detect defects.

Expected Outcomes


Hariharan Raveendran


Karan Rasathurai


Vilaxsan Vinasirajan



Dr. Isuru Nawinne

Senior Lecturer

HProf. Roshan G. Ragel


Keshara Weerasinghe

PhD Candidate