I need a bespoke machine-vision sofware that performs automatic quality inspection on finished manufactured products, flagging any surface flaws in real time as they pass the camera line. The workflow is already in place; what’s missing is a reliable, high-accuracy vision layer that can integrate with our existing conveyor-mounted GigE cameras and trigger the PLC when defects are found.
Core requirements
- Detect only surface-level defects (scratches, dents, stains, coating irregularities) with a minimum 95 % detection rate and less than 2 % false positives.
- Process 30 fps streams without dropping frames.
- Operate on Linux industrial PCs; GPU optimisation (CUDA/OpenCL) is welcome but not mandatory.
- Expose a REST or TCP socket so our MES can receive the pass/fail signal and a cropped defect image.
- Provide an intuitive calibration panel that lets technicians set camera gain, exposure and defect-size thresholds without touching code.
Suggested tech stack
OpenCV for image processing,
Python or C++ for the logic, and optionally TensorFlow/PyTorch if a CNN will outperform classic filters. I am open to your proposal as long as long-term licence costs stay low and source code is delivered.
Acceptance criteria
1. Demonstration video showing the system running on my sample product batch.
2. Executable installer (or Docker image) plus full commented source code.
3. User guide that covers installation, camera setup and tuning parameters.
4. A brief test report capturing detection rate, false positives and performance metrics on the supplied 500-image validation set.
Timeline about 1.5 months. I’m happy to answer any additional technical questions you may have before we start. Pay: ₹10,000.00 - ₹30,000.00 per month