I need a complete, ready-to-run video analytics tool that flags car-collision accidents in city-surveillance footage. The workflow must stay simple for the end user: they upload a clip, hit “Analyze,” and, within seconds, the screen returns something like: Accident Detected: YES Severity: HIGH (low | medium | high) Confidence: 92 % Time-stamp: 00:12 Key points to build in • Scope of detection: only car collisions; no pedestrian or bicycle tracking at this stage. • Source footage: city CCTV style feeds (fixed street-level cameras). • No real-time push notifications are required—the result can appear once processing is finished. I will rely on you to select or curate a robust, publicly available dataset (or a combination of datasets) that truly represents urban crash scenarios, then fine-tune an architecture such as YOLOv8, Faster R-CNN, or another proven model in PyTorch/TensorFlow. Accuracy and speed matter equally; anything under a few seconds for a one-minute clip on a modern GPU is ideal. Deliverables • Trained model files and all preprocessing scripts • Lightweight desktop or web demo (Python + OpenCV/Streamlit/Flask—your call) mirroring the UI flow above • At least three short demo videos that clearly show LOW, MEDIUM, and HIGH severity outputs • Brief setup guide so I can reproduce results locally or on a cloud VM This is time-sensitive, so please outline how quickly you can: 1. Finalize the dataset, 2. Train and validate the model, 3. Package the demo application. I also need a brief report on how the project was made.