Tutorial 8 — Traffic Analytics and Vehicle Counting with Jetson

Turn any camera overlooking a road or car park into an intelligent traffic analytics system. Count vehicles, classify them by type, generate flow heatmaps, and detect congestion — all in real time on the Jetson.

What you will learn

• How to detect and classify vehicles (car, truck, bus, motorbike, bicycle)
• How to count vehicles crossing a virtual line (tripwire counting)
• How to track individual vehicles across frames using ByteTrack
• How to generate positional heatmaps from traffic flow data
• How to export stats to CSV and visualise in Grafana

Step 1 — Run the traffic demo

cd ~/tutorials/08-traffic-analytics
python3 traffic.py --source 0 --show --count-line 400

A horizontal line appears at y=400 pixels. Every vehicle that crosses it is counted and classified. The running total is displayed in the corner.

Step 2 — Tripwire vehicle counting

from ultralytics import YOLO
from ultralytics.solutions import ObjectCounter

model   = YOLO("yolov8n.engine")
counter = ObjectCounter(
    names  = model.names,
    reg_pts= [(0, 400), (1280, 400)],   # horizontal counting line
    count_reg_color = (0, 255, 0)
)

while True:
    frame   = cap.read()[1]
    results = model.track(frame, persist=True)
    frame   = counter.start_counting(frame, results)
    cv2.imshow("Traffic", frame)

Step 3 — Generate a heatmap

import numpy as np

heatmap = np.zeros((720, 1280), dtype=np.float32)

def update_heatmap(detections):
    for box in detections:
        cx = int((box.xyxy[0][0] + box.xyxy[0][2]) / 2)
        cy = int((box.xyxy[0][1] + box.xyxy[0][3]) / 2)
        cv2.circle(heatmap, (cx, cy), 20, 1, -1)

# Normalise and overlay on frame
norm_heat = cv2.normalize(heatmap, None, 0, 255, cv2.NORM_MINMAX)
coloured  = cv2.applyColorMap(norm_heat.astype(np.uint8), cv2.COLORMAP_JET)

✅ Next: Tutorial 9 — Robot Vision | Back to Jetson Kit