Highway Violation Detection System

I need a real-time, camera-based solution that can watch highway traffic and automatically flag three key offences: riders without helmets, motorcycles carrying three people, and any vehicle that exceeds the posted speed limit. The goal is to integrate the model with existing roadside or CCTV cameras so that the moment a violation is detected, the system captures the frame, tags the offence type and time-stamp, and stores it in a database or pushes it to an API that my team will expose.

What I already have: several hours of highway footage in MP4 format, the speed-gun readings that correspond to those clips, and access to an NVIDIA GPU machine for training and inference.

What I expect from you:
• A trained, optimized computer-vision model (YOLOv8, RetinaNet, or another accurate architecture—open to your preference) that reaches at least 90 % precision on the three violations.
• An inference script (Python preferred) that ingests an RTSP stream, performs detection in real time, overlays bounding boxes and labels, and logs the events to a JSON file or REST endpoint.
• Clear instructions so that I can redeploy the solution on my own hardware or any cloud VM.

If you have previously worked with helmet detection, multi-rider classification, or speed estimation on highways, let me see the demos or repos. Accuracy, low latency, and straightforward deployment are my priorities, and I am ready to begin as soon as we agree on the approach.