Ultralytics YOLO Foundations: Part 2

Real World Use Cases & Solutions

2026-03-13

3. Real World Use Cases & Solutions

Beyond bounding boxes: actionable insights.

The “After” of Inference

You now know how to run tasks like detection and segmentation using pre-trained models.

But what happens AFTER the model outputs its predictions?

Often, you don’t just want bounding boxes drawn on an image.
You want actionable insights.
You want to trigger alerts, count items, or manage entire systems.

Ultralytics Solutions

Ultralytics provides a suite of high-level Solutions that act as wrappers around your AI model.

They allow you to deploy world-class computer vision systems with just a few lines of code.

Solution Categories

Region and Boundary Analytics: Counting, Zones, Parking, Queues
Kinematic and Spatial Measurement: Speed, Distance, VisionEye
Data Aggregation and Visualization: Analytics, Heatmaps
Image Output Manipulation: Cropping, Blurring
Specialized Model Heads: Instance Segmentation Tracking, Workouts
External Integrations: Live Inference (Streamlit), Security Alarms
Multimodal Embeddings: Similarity Search

Region & Boundary Analytics

Counting, Zones, and Queues

Region and Boundary Analytics

These solutions rely on defining geospatial boundaries (lines or polygons) in the frame and cross-referencing tracked object coordinates.

Object Counting: Evaluates line vector intersections.
Regions: Evaluates continuous occupancy inside a polygon.
Parking Management: Binary state (occupied/empty) for small polygons.
Queue Management: Measures object wait times in regions.
Track Objects in Zone: Masking filter to isolate tracking.

Example: Object Counting

Count objects passing through an area or crossing a line. View Official Documentation

import cv2
from ultralytics import solutions

cap = cv2.VideoCapture("videos/traffic.mp4")
line_points = [(100, 500), (900, 500)]

counter = solutions.ObjectCounter(
    show=True, region=line_points, model="yolov8n.pt"
)

while cap.isOpened():
    success, im0 = cap.read()
    if not success: break
    results = counter(im0)

Example: Parking Management

Organize and direct vehicle flow in parking areas. View Official Documentation

import cv2
from ultralytics import solutions

cap = cv2.VideoCapture("videos/traffic.mp4")
parking_spots = [
    [(10, 10), (10, 50), (100, 50), (100, 10)],
    [(110, 10), (110, 50), (200, 50), (200, 10)]
]

manager = solutions.ParkingManagement(
    model="yolov8n.pt", region=parking_spots, show=True
)

while cap.isOpened():
    success, im0 = cap.read()
    if not success: break
    results = manager(im0)

Kinematic and Spatial Measurement

Speed, Distance, and Proximity

Kinematic and Spatial Measurement

Calculate physical, real-world metrics based on relative 2D box positions.

Distance Calculation: Compares two moving objects to each other.
VisionEye View: Compares the distance from tracking objects to a single, static observer point.
Speed Estimation: Tracks the delta of an individual object’s position over time, accounting for perspective.

Example: Speed Estimation

Estimate object speed using tracking techniques and perspective transformations. View Official Documentation

import cv2
from ultralytics import solutions

cap = cv2.VideoCapture("videos/traffic.mp4")
# ... extract fps
speedestimator = solutions.SpeedEstimator(
    show=True, model="yolov8n.pt",
    fps=30, meter_per_pixel=0.05, 
)

while cap.isOpened():
    success, im0 = cap.read()
    if not success: break
    results = speedestimator(im0)

Data Aggregation & Visualization

Analytics and Heatmaps

Data Aggregation & Visualization

Aggregate detection and tracking data over time to produce high-level visual summaries and distributions.

Analytics: Plots statistical temporal trends (line graphs, pie charts) of class frequencies.
Heatmaps: Plots spatial density and accumulated movement paths directly onto the original image matrix.

Example: Analytics

Discover patterns and make informed decisions (Line, Bar, Area, Pie). View Official Documentation

import cv2
from ultralytics import solutions

cap = cv2.VideoCapture("videos/people.mp4")
analytics = solutions.Analytics(
    show=True,
    analytics_type="line", # "pie", "bar", "area"
    model="yolov8n.pt", 
)

frame_count = 0
while cap.isOpened():
    success, im0 = cap.read()
    if not success: break
    frame_count += 1
    results = analytics(im0, frame_count)

Image Output Manipulation

Cropping and Blurring

Image Output Manipulation

Act purely as downstream image processing steps utilizing box/mask coordinates.

Object Cropping: Slices out regions based on boxes and saves them as isolated image files (useful for dataset extraction).
Object Blurring: Applies a filter to obscure regions in the original full frame (privacy compliance).

Example: Object Blurring

Automatically blur bounding boxes or segmentation masks to protect privacy. View Official Documentation

import cv2
from ultralytics import solutions

cap = cv2.VideoCapture("videos/people.mp4")
blurrer = solutions.ObjectBlurrer(
    show=True,
    model="yolov8n.pt", 
    blur_ratio=0.5, # percentage of blur intensity
)

while cap.isOpened():
    success, im0 = cap.read()
    if not success: break
    results = blurrer(im0)

Specialized Model Heads

Segmentation and Pose

Specialized Model Heads

Utilize specific YOLO architectures (YOLO-seg, YOLO-pose) to extract granular, non-rectangular spatial data.

Instance Seg Tracking: Combines pixel-perfect masks with continuous track IDs.
Workouts Monitoring: Tracks human skeletal connections (pose keypoints) to evaluate biomechanical form.

Example: Workouts Monitoring

Track and evaluate exercise form (like pushups) using YOLO Pose keypoints. View Official Documentation

import cv2
from ultralytics import solutions

cap = cv2.VideoCapture("videos/people.mp4")
gym = solutions.AIGym(
    show=True,
    kpts=[6, 8, 10], # default keypoints for pushups
    model="yolov8n-pose.pt", 
)

while cap.isOpened():
    success, im0 = cap.read()
    if not success: break
    results = gym(im0)

External System Integrations

Alarms, Web Apps, Embeddings

External Integrations & Embeddings

Live Inference with Streamlit: Host a real-time, interactive web GUI for easy browser access.
Security Alarm System: Runs headless, communicates over SMTP to push email notifications upon triggers.
(Bonus) Similarity Search: Utilizes neural embeddings (CLIP) and Meta FAISS for natural language image retrieval.

Example: Security Alarm System

Trigger email alerts upon detecting specific objects. View Official Documentation

import cv2
from ultralytics import solutions

cap = cv2.VideoCapture("videos/people.mp4")
securityalarm = solutions.SecurityAlarm(
    show=True, model="yolov8n.pt",
    records=1, # send email after 1 detection
)

# Remember to use an App Password here if using Gmail!
securityalarm.authenticate("sen@email", "PASS", "rec@email") 

while cap.isOpened():
    success, im0 = cap.read()
    if not success: break
    results = securityalarm(im0)

Conclusion

Module Wrap-up

Summary

Solutions: High-level wrappers for tracking, counting, and geospatial analysis.
Analytics: Generating heatmaps and time-series data from detections.
Integration: Deploying security alarms and web demos.

Next Steps

Next up: Custom Data & Training

Learn how to prepare your own datasets for YOLO.
Train custom models to detect specific objects.
Understand hyperparameter tuning and model tracking.

Q&A

Thank You!

Any questions?