Worked on the huggingface/lerobot repository to enhance camera integration reliability on Windows by addressing pixel format enforcement in the OpenCVCamera module. Focused on the DSHOW backend, the developer fixed a bug where the OpenCV fourcc pixel format was not consistently applied after configuring resolution and FPS, which could lead to unpredictable video capture results. By restructuring the initialization sequence in Python and leveraging OpenCV, the fix ensures the requested pixel format is reliably enforced across devices. Inline documentation was added to clarify the ordering logic for future maintainers, resulting in improved stability and maintainability for Windows-based camera applications.