Computer Vision Developer — Automatic Basketball Tracking for Existing Flutter iOS App

I have an existing iOS app built in Flutter called Physics of the Shot. The app is already live on the App Store. The app allows a user to upload a basketball-shot video and calculates physics measurements such as launch angle, velocity, peak height and entry angle. Currently, the user manually marks key positions of the basketball in the video. For a potential future version, I am exploring whether this process could be improved through automatic basketball detection and tracking. The goal would be for the app to: 1. Detect the basketball in an uploaded video. 2. Track the center of the ball across successive video frames. 3. Produce reliable x/y coordinates and timestamps for the detected ball positions. 4. Use those points to reconstruct the ball’s trajectory. 5. Pass the trajectory data into the app’s existing physics/calculation system. 6. Ideally allow the user to review or correct the automatically detected trajectory if detection is imperfect. This is not a request to rebuild the existing app. The main task is to investigate and potentially implement the computer-vision/tracking component within the existing Flutter/iOS architecture. I am open to the appropriate technical approach — for example YOLO, Core ML, TensorFlow Lite or another suitable object-detection/tracking solution. I am more interested in accuracy, reliability and practical on-device performance than in using a particular model. Videos may include real-world challenges such as: * different basketball sizes on screen * outdoor and indoor courts * changing lighting/backgrounds * players temporarily obscuring the ball * camera movement * different frame rates, including slow-motion footage Ideal experience: * Computer vision / object detection * Object tracking across video frames * YOLO or comparable detection models * Core ML and/or TensorFlow Lite * Flutter integration * iOS video/frame processing * Experience deploying ML models on-device rather than only running them in Python/server environments 1. Have you previously built a system that detects and tracks an object across video frames? Please provide an example if possible. 2. What technical approach would you recommend for automatic basketball detection and tracking in this app, and why? 3. Would you recommend running the detection entirely on-device (iPhone) or server-side? 4. How would you handle missed detections/occlusion and turn frame-by-frame detections into a smooth, reliable ball trajectory? 5. Roughly how many hours would you estimate for an initial proof of concept using a small set of basketball-shot videos?