Real-Time Android Eye Cropper
Budget: €250 – €750 EUR
I’m building an Android app for general eye-tracking and need a standalone capture module that records only the ocular strip in real time. MediaPipe Face Mesh should locate the eyes, and everything outside that region must be discarded before any frame is written to disk so no full-face data ever leaves memory.
Although the front camera is the primary lens we will use, the code should still be able to switch to the rear unit or a UVC camera when connected. Smoothness is critical; I want a consistent 30–60 fps even during rapid saccades, so please optimise the processing pipeline with OpenCV, GPU acceleration, or other techniques you prefer.
The finished module must:
• take a live camera feed,
• crop it in real time to the eye-strip,
• save the result as .mp4, and
• drop into my existing project with minimal wiring.
Please include the full Android Studio project, clearly commented Kotlin or Java source, and a brief README explaining any build flags, permissions, or MediaPipe versions used. I’ll review performance on a Pixel device first, then on a tablet with an external UVC camera; meeting the frame-rate target under both scenarios will be the key acceptance criterion.
Although the front camera is the primary lens we will use, the code should still be able to switch to the rear unit or a UVC camera when connected. Smoothness is critical; I want a consistent 30–60 fps even during rapid saccades, so please optimise the processing pipeline with OpenCV, GPU acceleration, or other techniques you prefer.
The finished module must:
• take a live camera feed,
• crop it in real time to the eye-strip,
• save the result as .mp4, and
• drop into my existing project with minimal wiring.
Please include the full Android Studio project, clearly commented Kotlin or Java source, and a brief README explaining any build flags, permissions, or MediaPipe versions used. I’ll review performance on a Pixel device first, then on a tablet with an external UVC camera; meeting the frame-rate target under both scenarios will be the key acceptance criterion.