SIGNAL-IQ: AN INTEGRATED, LOW-COST, HARDWARE-VALIDATED ADAPTIVE TRAFFIC SIGNAL CONTROL SYSTEM USING YOLOV8-BASED VEHICLE AND EMERGENCY-VEHICLE DETECTION

Abstract

Fixed-time controllers run most intersections in Pakistan and in comparable settings. They hand an approach its full green whether a queue waits behind the stop line or the lane sits empty. Infrastructure-driven adaptive platforms such as SCATS and SCOOT correct this, and they depend on detector networks whose installation and upkeep cost more than many municipal budgets allow. A large share of published reinforcement-learning controllers stops at simulation. Signal-IQ is an adaptive controller assembled entirely from off-the-shelf parts: a webcam feed, two separately trained YOLOv8 detectors, a rule-based Python controller, and an Arduino Uno R3 driving twelve LEDs arranged as four three-lamp signal heads. The controller counts vehicles inside four per-approach regions of interest, sets the green interval from the count, skips approaches with no vehicles, and hands an approach an unconditional green when the second detector reports an ambulance, fire truck, or police car. Seven use cases fix the functional scope across surveillance, detection, direction-based counting, and emergency override. A documented suite of ten test cases exercises the models, the control code, the Arduino firmware, the integrated pipeline, and four traffic scenarios, plus actuation on physical LED signal heads. On a held-out test set the general vehicle model (car, bus, truck, motorcycle) reached 0.94 precision, 0.91 recall, and 0.93 mAP@0.5. The emergency-vehicle model (ambulance, fire truck, police car) reached 0.98, 0.96, and 0.97. Measured against a fixed-time baseline with equal 30-second splits, average vehicle wait time across three simulated scenarios fell by 72.9% to 81.8%. The contribution is integration and physical validation rather than a new detector or a new control algorithm. We state the evaluation work still undone directly: comparison against a published adaptive controller, ablation of the individual control features, repeated trials with reported variance, and field deployment.