Python Developer Needed for OpenCV Template Matching (Multi-Resolution Screens)

We are looking for a Computer Vision engineer with strong Python and OpenCV experience to optimize an existing image-processing pipeline. Our current implementation relies on OCR, which is computationally expensive and limits throughput, also we have commented code which does the template matching for some resolutions @ 100+ image/sec. Our goal is to replace OCR usage by introducing efficient template matching, allowing us to process more images per second with lower latency and cost. Attached are: 2 example screenshots where the system must detect the “Kickoff” UI element 1 example screenshot where the system must detect the “Final” UI element These are only example references to define what needs to be detected. You must source or generate additional Madden 2026 screenshots across multiple screen resolutions and aspect ratios (e.g. 1080p, 1440p, 4K, ultrawide) to validate that detection works consistently. Your solution must correctly handle both: Positive cases: If “Kickoff” / “Final” is present, return the match location (x, y, w, h) and a confidence score. Negative cases: If the template is not present, return null / mismatch (no false positives). We care about high precision (avoid false positives), and we can optionally fall back to OCR only if confidence is low (but the goal is to reduce OCR). We want coverage for: Different resolutions / scaling Slight UI shifts (positions change) Compression differences (if any) Different match situations where “Kickoff” / “Final” appears The API is already implemented and returns detection results. Your task is only to plug in your OpenCV template-matching logic. No API design, no backend refactor, no UI work is required. You will integrate your code into the existing Python pipeline. Test Coverage & Validation Script (Mandatory) You must provide a standalone Python test script that: Runs the detection on a folder of images Covers: Positive cases (Kickoff / Final present) Negative cases (neither present) Outputs: Match found → coordinates + confidence No match → null / mismatch Passes all provided positive and negative test cases Demonstrates zero false positives on negative images Please share similar projects and your proposed approach.