Adaptive Traffic Signal Management (YOLOv8 + SORT) — Vehicle Counting, Line Crossing, Adaptive Timing

Demo

ATSM demo — YOLOv8 vehicle counting and adaptive timing

Quickstart

py -3.12 -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt
python car_counter.py

Download ZIP

How it works

YOLOv8 detects vehicles in the ROI.
SORT (Kalman + IOU) maintains stable IDs across frames.
Line crossing increments a unique vehicle count.
Red timer reduces when congestion exceeds a threshold.

Configuration and CLI

Use config.yaml for defaults and override via CLI:

# Default
python car_counter.py

# Custom config
python car_counter.py --config my_config.yaml

# Override key params
python car_counter.py --video Media/cars3.mp4 --weights Weights/yolov8l.pt \
  --line 150 360 1200 360 --threshold 15 --normal 70 --reduced 40 --cooldown 15 --conf 0.4

Detailed explanation

Detection

Ultralytics YOLOv8 runs on the masked region-of-interest to reduce false positives and compute cost.

Weights: Weights/yolov8n.pt (fast) or yolov8l.pt (more accurate).
Classes counted: car, truck, motorbike, bus.
Confidence threshold configurable via --conf.

Tracking (SORT)

Kalman filter predicts positions; IOU-based matching assigns detections to tracks with unique IDs.

Key params: max_age, min_hits, iou_threshold.
Stable IDs prevent double-counting across adjacent frames.

Counting

When a tracked center crosses the configured line window, the ID is marked counted once per cycle.

Line coordinates: count_line = [x1, y1, x2, y2].
Change via config or --line x1 y1 x2 y2.

Adaptive timing

If the unique count exceeds the threshold, the red timer reduces for the current cycle with a simple cooldown to prevent oscillation.

Threshold: car_count_threshold.
Timers: normal_red_timer, reduced_red_timer.
Cooldown: cooldown_duration.

Tips & troubleshooting

Use yolov8n.pt for CPU or low-spec GPUs; upgrade to yolov8l.pt for accuracy checks.
Adjust --conf if you see false positives/negatives.
Tune SORT params for your traffic scene density/speed.
Replace Media/mask.png for a new intersection layout.

Cite this work

Please cite the repository if you use or build upon it.

DOI: 10.5281/zenodo.16903140

@software{jambhale_atsm_2025,
  author  = {Jambhale, Anuj},
  title   = {Adaptive Traffic Signal Management (ATSM): Vehicle Counting with YOLOv8 + SORT},
  year    = {2025},
  url     = {https://github.com/jambhaleAnuj/Traffic_signal_counter_using_car_count_python},
  doi     = {10.5281/zenodo.16903140}
}