CCTV-TO-ROUTE INTELLIGENCE FOR REAL-TIME AMBULANCE MOVEMENT, TRAFFIC OBSTRUCTION DETECTION AND EMERGENCY CORRIDOR OPTIMIZATION - (CLEAR PATH EMS)

Abstract

The time taken by an ambulance to reach a patient and then a hospital often decides the outcome in cardiac arrest, stroke, trauma, and obstetric emergencies. In congested cities, a large part of this time is lost in traffic queues at signals, junctions blocked by stalled vehicles, and roads narrowed by construction or illegal parking. Traffic police and ambulance drivers usually learn about these obstacles only when they reach them. Meanwhile, many cities have installed hundreds of CCTV cameras at junctions, whose video is watched manually in control rooms. This paper presents Clear Path EMS, a system that converts CCTV video into route intelligence for real-time ambulance movement tracking, traffic obstruction detection, and emergency corridor optimisation. The system processes video streams from junction cameras using a YOLOv8 object detection model to count and classify vehicles, estimate queue length, and detect obstructions such as stalled vehicles, accidents, and blocked lanes. Detected ambulances are tracked across frames with the DeepSORT algorithm, and their positions are combined with GPS data sent from the ambulance. Rather than storing raw video, the edge processing nodes publish compact events such as vehicle counts, queue lengths, and obstruction alerts, which flow through a message stream into a time-series store and a road network graph. The route intelligence layer converts these events into live travel time estimates for each road segment. A gradient boosting model predicts segment travel times for the next fifteen minutes using current counts, queue lengths, historical patterns, time of day, and weather. A modified shortest path algorithm on the road graph then selects the fastest route for the ambulance, avoiding segments with obstructions, and identifies the signals along the route that should be given a green wave. The traffic control room receives a corridor plan listing junctions, expected arrival times, and actions required at each point. A control room dashboard shows ambulance positions, the planned route, cameradetected congestion, active obstructions, and signal priority status, while the ambulance crew receives turn-by-turn guidance. In an evaluation using recorded video from twenty junctions and replayed ambulance trips, vehicle detection achieved a mean average precision of 0.89, obstruction alerts were raised within about 20 seconds, and the optimised corridors reduced simulated travel time by about 27 percent compared with routes chosen by conventional navigation.