AI/Computer Vision Developer for Traffic Video Analysis (YOLO + Trajectory Tracking)

AI/Computer Vision Developer for Traffic Video Analysis (YOLO + Trajectory Tracking)

AI/Computer Vision Developer for Traffic Video Analysis (YOLO + Trajectory Tracking)

Upwork

Upwork

Remote

2 hours ago

No application

About

Main Task: Develop a reusable, modular Python-based system to analyze traffic videos by detecting, tracking, and classifying road users (vehicles and pedestrians), recognizing license plates, estimating vehicle occupancy, and optionally performing facial recognition for visible passengers and pedestrians. The system must generate annotated video outputs and structured data logs (CSV/Excel) for further analysis. Goal: The system should support object detection for multiple road user types, including cars, buses, trucks, bicycles, motorcycles, and pedestrians, and apply multi-object tracking to assign consistent IDs across frames. It should classify movements at intersections (e.g., left turn, right turn, through, U-turn), and incorporate license plate recognition using OCR-based methods. Additionally, the system should estimate vehicle occupancy when possible, using in-vehicle detection techniques to count visible passengers, particularly in front seats. The system needs to include facial recognition or face detection to identify or re-identify pedestrians and vehicle passengers. All outputs should include annotated videos with bounding boxes, object IDs, trajectories, license plate overlays, and face detections when applicable. Structured logs (CSV/Excel) must be generated to record object classes, movements, counts, and timestamps. Based on the extracted data, the system should compute essential traffic engineering metrics, including turning movement counts, vehicle classification counts, peak hour volumes, and modal split. It should also calculate queue lengths, average delay per vehicle, saturation flow rates, vehicle speeds, and headways. Additional analyses may include conflict detection (e.g., rear-end, lane-change, crossing conflicts), pedestrian and cyclist interactions, and generation of trajectory heatmaps. All metrics rely on established formulas, and we will provide support for their implementation. The final deliverables must include a clean, modular, and well-documented Python codebase, example video and data outputs, and a clear setup and user guide. Please see the attached file, it shows an example of what we are looking for. Thanks,