CV supported defective pixel detector tool
Services:
Project overview
Our client is a leader in designing LED technology who delivers digital displays to customers across industries. Sports events, roadside signs, school scoreboards, commercial advertising – the well-known US manufacturer has been in the LED game since the late ’70s.
Their in-house testing facility is conducting rigorous assessments to ensure product reliability and longevity. However, even market leaders are facing technical challenges.
With the critical need for flawless daily operation, the client has sought to facilitate its usual quality control. Given its brand image, all-manual workflows were no longer justified.
That’s where Abto Software has entered the partnership to design a solution for automated visual inspection by leveraging computer vision image analysis.
Main goals
The primarily project objective was focused around replacing the inefficient manual workflows of assessment. Before that, the operators heavily relied on rigorously, manually marking faulty components in spreadsheets – a resource-intense, error-prone approach.
We aimed at streamlining existing workflows by automating:
- The process of detecting faulty diodes
- And reporting of found LED malfunctions
Our contribution
Phase 1: The proof-of-concept
At this project stage, we started by processing a folder of photos – raw images of to-be-inspected LED panels.
In brief:
- We built an algorithm to identify faulty diodes
- And proved approach viability
Phase 2: User-friendly design, workflow automation, and optimization
At the next stage, we designed and created a program for operators to fine-tune the parameters for inspection, highlight defects more accurately, and adjust the radius of detection.
To short:
- We developed an intuitive user interface
- And expanded the algorithm to process more types of panels
What’s more, we introduced the generation of automated Excel reports detailing locations and coordinates, thereby making the solution more functional and ready for applied, real-world utilization.
We covered:
- UI design & development
- Codebase cleanup & optimization
- Application framework
- Algorithm enhancement & adjustments
- UI integration & automation
- UI enhancement, bug fixes, and testing
Phase 3: Final amendments & extension
At the final stage, we implemented security keys to encrypt the application and prevent unauthorized access, afterwards adding additional features.
We introduced:
- Color detection (blue, green, red, cyan, magenta, yellow, and black) and standardized marker colors
- And cursor-based coordinate tracking to locate faulty diodes with accuracy
How the solution works
Our tool is designed to automate LED panel defect detection to replace the process of marking faulty diodes and streamline quality control by leveraging computer vision.
The concept:
Image processing
- The operator first uploads the images of the LED panels
- The optical inspection system then scans the images, accurately detects and highlights faulty diodes
Custom parameters
- The operator can fine-tune the settings by specifying panel width and height, pitch size, and layout
- The additional options include:
– Background blurring
– Point radius adjustment
– Grid size modification
Automated reporting and tracking
- The spot-on LED positions can be then tracked in real-time with cursor-based coordinate display
- Our optical inspection system will generate a detailed Excel report with precise LED coordinates
Advanced features and security
- The dead pixel detector can detect multiple colors to improve the analysis
- The defect pixel detector is protected by implemented security keys to prevent unauthorized access
Main challenges
Real-world robustness
The images being uploaded are typically low quality: glare, blurriness, poor lighting or perspectives, and more. That means limited consistency in detecting faulty diodes.
We addressed this problem by leveraging preprocessing techniques – image cleansing, alignment adjustments, and filters for accurate quality control.
Future scalability
New models being introduced with regularity and velocity are creating the need for adaptability and scalability. A completely manual adjustment would cause time- and cost-related overload.
We resolved this problem by building the algorithm to be highly flexible and modular:
- Universal parameters are designed to handle common scenarios
- Configurable parameters are intended to allow easy adaptation to new LED types
Tools and technologies:
AI/ML & tech stack:
- Python
- OpenCV
Libraries & tools used:
- PyQt
- PyArmor
- SciPy
Timeline:
- July 2023 – March 2024
Team:
- 1 project manager
- 1 CV engineer
- 1 Python developer
- 1 QA engineer
Value delivered to business
LED displays are intricate and require thorough testing – from components to completely assembled systems. Any misfires can inflict substantial consequences – reputational and financial losses, and, what’s more critical, safety risks in different high-stake environments.
By delivering computer vision image analysis, we enabled:
Reduced time-to-market
More efficient quality control can improve production cycles and enhance product rollout, which translates:
- Into greater product offering
- And stronger market positioning
Reduced costs
An automated defect detection can minimize manual workflows and associated human error, which means:
- Sensibly lower labor expenses
- And fewer product withdrawals
Higher quality and reliability
With automated defect detection, the company can ensure only high-quality LED panels will reach the market. That means reduced returns, warranty claims, customer frustration, and, accordingly, reputational damage.
Data-driven decision-making
With automated defect reporting & tracking, the company can harness valuable insights into trends and issues. That means thought-out strategies and potentially bigger revenue.