Web Chicken Crossing Predictor

I need a small web-based bot that looks at a live camera feed of a country road and tells me, with reasonable confidence, whether a chicken is about to step onto the asphalt within the next few seconds. The model should run in real time, push its verdict to the page instantly, and refresh continuously while the stream is active.

Here is the flow I have in mind. A browser tab shows the video; your back-end service—Python with OpenCV or a similar CV library is fine—grabs each frame, applies the trained model, and returns a simple “Likely to cross / Unlikely” flag plus a probability score. I am comfortable hosting on a basic VPS, so please keep deployment lightweight (Docker container or similar). If TensorFlow, PyTorch, or YOLO is your preferred stack, go for it.

I am supplying only the live feed; there is no archive of labelled clips, so you will need to propose a quick data-gathering or augmentation strategy to bootstrap training, or leverage transfer learning from an existing object-detection network. Accuracy matters more than aesthetics, but a clean HTML page that shows the stream and the prediction overlay will help me demo the project to stakeholders.

Deliverables
• A trained model capable of predicting imminent chicken crossings from the live stream
• Source code for both the inference service and the minimal web front end
• Deployment script or Dockerfile with clear, one-step setup instructions
• Brief read-me summarising how to retrain or fine-tune the model as new footage becomes available

I will test the solution by pointing it at my camera for a full day and checking that alerts correlate with actual crossings. Let me know any extra resources you might need and your rough timeline.